An adaptive modified Newton–Raphson-based optimizer for efficient optimal power flow performance under multiple operating conditions Authors: Ali S. Aljumah, Mohammed H. Alqahtani, Abdullah M. Shaheen, Mohamed O. Atallah Published: 2026 ABSTRACT The optimal power flow (OPF) problem is essentially about finding the cheapest and safest way to operate a power system without breaking any of the operational limits that govern it. In this paper, we introduce a new Modified Newton–Raphson-Based Optimizer (MNRBO) specifically designed to tackle real-world OPF problems, integrating renewable photovoltaic sources. The NRBO integrates gradient-inspired search using the NR search rule and the trap avoidance strategy. Our MNRBO extends this framework by adding two adaptive components. An Adaptive Crossover Mechanism (ACM) is added that lets solutions dynamically exchange useful information with each other, keeping the population diverse and preventing everyone from getting stuck in the same mediocre spot too soon. Also, a Sigmoid decay mode that smoothly and gradually shifts the algorithm from broad exploration (looking around the whole search space) in the early stages to careful fine-tuning (exploitation) toward the end. This gives much steadier and more predictable convergence than the original abrupt or polynomial decay. The resulting MNRBO algorithm forms a self-evolving optimization framework that automatically adjusts its learning strategy as the search progresses. We thoroughly tested MNRBO on the standard IEEE 30-bus system across a wide range of realistic scenarios: minimizing fuel costs (with smooth quadratic models, valve-point ripples, and multi-fuel options), handling generators with prohibited operating zones, and minimizing transmission losses under normal, peak, and light-load conditions. In every single case, MNRBO delivered better solutions, faster and more consistent convergence, and dramatically lower variation across multiple runs compared to the original NRBO and several other state-of-the-art algorithms. The results clearly show that MNRBO is not only more accurate but also far more robust and dependable, exactly what operators need when solving OPF in real power systems where reliability really matters. To further validate the applicability of the proposed approach under renewable energy uncertainty, a probabilistic OPF framework incorporating photovoltaic renewable generation is developed. In this case study, the integration of renewable solar photovoltaic energy in conditions of variable irradiance is examined using the Point Estimate Method (PEM) with lognormal irradiance modeling. In addition, an ablation study is conducted to quantify the individual contributions of the ACM and sigmoid decay strategy in the presence of renewable photovoltaic sources, demonstrating their significant impact on convergence stability, robustness, and optimization accuracy. Introduction The Optimal Power Flow (OPF) problem remains one of the most fundamental and challenging tasks in modern power system operation and planning. At its core, it involves determining the optimal values of control variables, including real and reactive power outputs of generators, voltage magnitudes at generator buses, transformer tap settings, and shunt compensator outputs1, one or more objective functions such as fuel cost, power losses, voltage deviation, or emissions. This optimization process must simultaneously satisfy the system equality constraints represented by power balance equations as well as inequality constraints including generator limits, bus voltage limits, transmission line capacities, and security requirements2. It has long been recognized as a key optimization task in modern power system operation, introduced initially by Carpentier in the 1960s and later formalized by Dommel and Tinney3. The nonlinear, nonconvex, and highly constrained nature of OPF makes it difficult to solve using classical mathematical optimization techniques4,5, especially under multiple and uncertain operating scenarios. Because of that nasty mathematical landscape, traditional gradient-based methods, interior-point, sequential quadratic programming, and the like, often get trapped in local optima or simply fail to converge reliably, especially when the system is large, heavily loaded, or facing uncertainties. The problem is inherently complex, particularly when incorporating renewable energy sources (RES) with their inherent uncertainty and intermittency, making classical gradient-based methods prone to local optima or convergence issues. Consequently, recent research has heavily emphasized metaheuristic and nature-inspired optimization algorithms, which offer robust global search capabilities, faster convergence, and the ability to handle multi-objective formulations and large-scale systems6,7. Recent research emphasizes bio-inspired and physics-inspired metaheuristics for applications such as reactive power compensation, renewable integration, and economic-environmental dispatch. In8, an updated hummingbird algorithm was used to optimise assigning of Static VAR Compensators (SVCs) in distribution systems, with an emphasis on maximising energy efficiency. This upgrade helped with the installation of FACTS devices for loss minimisation and voltage profile augmentation. In a related study, a modified artificial rabbits optimizer was designed for both siting and dynamic operation of Distribution-SVCs9, adapting to varying load conditions. Modifications to the original rabbit optimiser included features of increased mutation and opposition-based learning to boost performance in power-constrained systems. These technologies demonstrated reliable solutions for real-time control in distribution systems. In10, an enhanced fractional order Kepler optimizer was developed to efficiently integrate STATCOM devices to tackle multi-modal challenges in OPF with renewable PV sources. In this work, improving energy efficiency was addressed by minimizing losses or costs, maximizing PV hosting capacity, and ensuring stability in hybrid PV-STATCOM configurations. For large-scale systems, a weighted mean of vectors technique was developed for Combined Economic and Environmental Dispatch (CEED)11. In this work, the multi-objective problem of fuel costs and emissions was balanced, utilizing vector combining, rule updating, and local searching operators in solving large-scale OPF variants, focusing on scalability and constraint management. While the primary focus is on power flow optimization, complementary works apply data-driven and optimization techniques to electric vehicle (EV) and battery systems, which interact with power grids. For instance, studies on lithium-ion battery fault diagnosis12, state-of-health estimation13, retirement prediction14, and energy consumption analysis of electric buses15 support broader smart grid and vehicle-to-grid (V2G) optimization frameworks. These can feed into OPF models by providing accurate load or storage representations, though they primarily employ machine learning rather than metaheuristics for core optimization. Over the past two decades, the power systems community has increasingly turned to metaheuristic algorithms16. These methods don’t need derivatives, they’re remarkably flexible, and they’re designed from the ground up to hunt for good (sometimes near-global) solutions even in the messiest search spaces. Among the earliest and most popular are the Particle Swarm Optimizer (PSO) and17, Differential Evolution (DE) algorithm18, DE with spherical for search space partitioning19, PSO20, colliding bodies optimization algorithm21, Sine-Cosine algorithm22, grey wolf optimizer23, and Zebra optimisation algorithm24. These algorithms mimic natural or social phenomena to explore the solution space effectively and have been successfully applied to OPF for objectives such as real power loss minimization, fuel costs minimization, and total voltage deviation. Despite their success, many of these algorithms face issues such as premature convergence and slow exploitation capability. Consequently, various enhanced or hybrid variants have been proposed. In25, the ensemble of constraint handling techniques-DE integrated multiple constraint-handling methods to efficiently address OPF’s complex operational limits. In26, a modified Jellyfish method incorporating a quasi-reflection learning strategy was presented to solve the OPF issue. In this approach, each jellyfish randomly selected a cluster from the population to represent its social group for information sharing and validated using the IEEE standard systems beside a practical network region in Egypt. A growing body of work from 2023 to 2025 highlighted the effectiveness of advanced metaheuristics tailored for OPF. The whale migration algorithm, as demonstrated by Ghasemi et al. (2024), effectively modeled migration behaviors to enhance fuel cost and power loss minimization in power systems while adhering to operational constraints27. Additionally, Ghasemi et al. (2023) presented the Gaussian bare-bones Levy circulatory system-based optimization (GBLCSBO) for optimizing power flow with renewable energy sources28. In this work, the CSBO algorithm’s capacity to handle the stochastic nature of wind and solar generation was improved by incorporating Gaussian bare-bones sampling and Levy flight distributions within the circulatory system framework. Further advancements in comparative metaheuristic analysis were provided by Diab et al. (2024). In this study, six recent algorithms were evaluated, including Salp Swarm Algorithm (SSA), Gradient-Based Optimizer (GBO), Northern Goshawk Optimization (NGO), fast cuckoo search, Dynamic Control Cuckoo Search (DCCS), and Opposition Flow Direction Algorithm (OFDA)—across multiple OPF objectives on IEEE 30-bus and 118-bus systems6. The study found that GBO consistently outperformed in optimality, convergence speed, and statistical robustness across both single- and multi-objective scenarios, including fuel cost minimization, voltage profile improvement, voltage stability enhancement via L-index, and loss minimization6. Also, Aurangzeb et al. (2023) introduced a hybrid metaheuristic approach, by integrating the bird swarm algorithm and jaya optimizer for OPF analysis. This method was evaluated on standard benchmarks, showing significant performance in minimizing generation costs, emissions, voltage deviations, and line losses compared to standalone algorithms29. Most recently, Nassef et al. (2025) applied the red-tailed hawk optimization algorithm to OPF, demonstrating enhanced performance through its hunting-inspired search mechanisms30. In this work, the employed red-tailed hawk optimization demenstrated lower objective values and improved stability margins compared to established techniques, especially in systems with varying load and contingency conditions. Other notable improvements include the Quasi-Reflection-Based Symbiotic Organisms Search (QR-SOS)31, bird swarm technique with orthogonal learning strategy32, enhanced Kepler optimizer33, a composite DE approach with different constrained handling strategies34, and a modified social network algorithm35, each designed to enhance global exploration and stability in constrained optimization scenarios. These studies illustrate a clear trend toward hybrid, Levy-enhanced, and bio-inspired metaheuristics that overcome the limitations of traditional solvers. On the other side, the cases dealing with hydro-thermal coordination modelled within the OPF framework typically extend the classical OPF formulation to jointly optimize the scheduling of hydroelectric and thermal generating units over a planning horizon. These models incorporate additional constraints36 such as water balance equations, reservoir storage limits, inflow forecasts, and hydro generation curves alongside standard power flow and operational limits, transforming the problem into a coupled hydro-thermal scheduling and OPF task. Metaheuristic approaches, such as the sine cosine algorithm applied to the IEEE 39-bus system, have been employed to solve this integrated formulation by minimizing total operating costs, including thermal fuel and hydro opportunity costs, while ensuring load-generation balance and transmission security under varying demand profiles37. Such coordinated models account for the spatiotemporal dynamics between flexible hydro resources and dispatchable thermal plants, often yielding significant reductions in fuel consumption, improved voltage profiles, and enhanced system reliability compared to decoupled solutions. The evolution of OPF-solving methodologies has shifted from deterministic mathematical programming to adaptive, self-evolving optimization. Metaheuristic approaches are preferred due to their effectiveness in tackling large-scale, non-linear, and multi-objective problems that classical techniques struggle with38. These methods balance exploration of new solution areas with refining existing solutions39 and are widely applicable in engineering40. Each new algorithm aims to enhance the balance between global and local search behaviors while addressing the operational constraints of real-world power systems. A recently created population-based metaheuristic algorithm of NRBO was given by R. Sowmya et al.41 with succesful implementations in a variety of domains. The NRBO algorithm was succesfully utilized in solving several engineeering problems such as antenna array beam synthesis in wireless communications42, parameter identification of solar PV systems43, integration of fluid antenna systems and reconfigurable surfaces in wireless systems44, control of fuel cells45, and grid-forming with converter control in wind power networks46. In this paper, we take NRBO further and present a Modified Newton–Raphson-Based Optimizer (MNRBO). We maintain NRBO’s core strengths, including the Newton–Raphson search rule and trap avoidance strategy, while introducing two adaptive enhancements that significantly improve performance: an Adaptive Crossover Mechanism (ACM) that lets solutions intelligently share good traits with each other, keeping the population diverse and fighting premature convergence, and.a smooth Sigmoid decay mode that gently steers the algorithm from wide-ranging exploration at the start to precise exploitation near the end. an Adaptive Crossover Mechanism (ACM) that lets solutions intelligently share good traits with each other, keeping the population diverse and fighting premature convergence, and. an Adaptive Crossover Mechanism (ACM) that lets solutions intelligently share good traits with each other, keeping the population diverse and fighting premature convergence, and. a smooth Sigmoid decay mode that gently steers the algorithm from wide-ranging exploration at the start to precise exploitation near the end. a smooth Sigmoid decay mode that gently steers the algorithm from wide-ranging exploration at the start to precise exploitation near the end. The main contributions of this study are therefore: A novel population-driven variant of NRBO is proposed, integrating Newtonian local refinement with adaptive evolutionary learning for improved OPF optimization.An ACM is introduced to promote dynamic knowledge exchange, enhances diversity, and prevents premature convergence, while a Sigmoid-based decay function ensures smooth transition between exploration and exploitation phases.The MNRBO is applied to seven OPF cases on the IEEE 30-bus system, demonstrating its ability to minimize generation cost, reduce losses, and maintain feasibility under varying load and contingency conditions.Statistical analyses and boxplots reveal that MNRBO offers the highest stability among all tested algorithms, exhibiting minimal spread and consistent convergence across repetitions.A probabilistic OPF framework considering uncertain solar PV generation is incorporated using the PEM, enabling the proposed MNRBO to efficiently handle renewables under stochastic operating conditions.The stochastic behavior of solar irradiance is modeled using a lognormal PDF, and multiple PEM scenarios are utilized to evaluate the robustness of the proposed MNRBO in probabilistic QFC minimization problems.An extensive ablation analysis is additionally performed to separately investigate the effects of the ACM and the Sigmoid Decay strategy on the overall optimization performance.The proposed MNRBO is further validated on both probabilistic renewable-integrated IEEE 30-bus and large-scale IEEE 118-bus systems, confirming its scalability, robustness, and convergence reliability for high-dimensional OPF problems. A novel population-driven variant of NRBO is proposed, integrating Newtonian local refinement with adaptive evolutionary learning for improved OPF optimization. A novel population-driven variant of NRBO is proposed, integrating Newtonian local refinement with adaptive evolutionary learning for improved OPF optimization. An ACM is introduced to promote dynamic knowledge exchange, enhances diversity, and prevents premature convergence, while a Sigmoid-based decay function ensures smooth transition between exploration and exploitation phases. An ACM is introduced to promote dynamic knowledge exchange, enhances diversity, and prevents premature convergence, while a Sigmoid-based decay function ensures smooth transition between exploration and exploitation phases. The MNRBO is applied to seven OPF cases on the IEEE 30-bus system, demonstrating its ability to minimize generation cost, reduce losses, and maintain feasibility under varying load and contingency conditions. The MNRBO is applied to seven OPF cases on the IEEE 30-bus system, demonstrating its ability to minimize generation cost, reduce losses, and maintain feasibility under varying load and contingency conditions. Statistical analyses and boxplots reveal that MNRBO offers the highest stability among all tested algorithms, exhibiting minimal spread and consistent convergence across repetitions. Statistical analyses and boxplots reveal that MNRBO offers the highest stability among all tested algorithms, exhibiting minimal spread and consistent convergence across repetitions. A probabilistic OPF framework considering uncertain solar PV generation is incorporated using the PEM, enabling the proposed MNRBO to efficiently handle renewables under stochastic operating conditions. A probabilistic OPF framework considering uncertain solar PV generation is incorporated using the PEM, enabling the proposed MNRBO to efficiently handle renewables under stochastic operating conditions. The stochastic behavior of solar irradiance is modeled using a lognormal PDF, and multiple PEM scenarios are utilized to evaluate the robustness of the proposed MNRBO in probabilistic QFC minimization problems. The stochastic behavior of solar irradiance is modeled using a lognormal PDF, and multiple PEM scenarios are utilized to evaluate the robustness of the proposed MNRBO in probabilistic QFC minimization problems. An extensive ablation analysis is additionally performed to separately investigate the effects of the ACM and the Sigmoid Decay strategy on the overall optimization performance. An extensive ablation analysis is additionally performed to separately investigate the effects of the ACM and the Sigmoid Decay strategy on the overall optimization performance. The proposed MNRBO is further validated on both probabilistic renewable-integrated IEEE 30-bus and large-scale IEEE 118-bus systems, confirming its scalability, robustness, and convergence reliability for high-dimensional OPF problems. The proposed MNRBO is further validated on both probabilistic renewable-integrated IEEE 30-bus and large-scale IEEE 118-bus systems, confirming its scalability, robustness, and convergence reliability for high-dimensional OPF problems. The present study considers a generation portfolio composed exclusively of thermal units, as defined in the standard IEEE 30-bus test system adopted in this work. This consideration reflects a widely used and practically validated benchmark in OPF research, where all generating units are modeled with thermal characteristics, including quadratic, valve-point, and multi-fuel cost functions. Although hydro-thermal coordination represents another important extension in power system optimization, it introduces additional complexities such as water balance constraints, reservoir dynamics, and time-coupled scheduling, which are beyond the scope of the current study. The primary objective of this work is to rigorously evaluate the effectiveness of the proposed MNRBO algorithm under highly nonlinear and constrained thermal OPF scenarios. Therefore, focusing on thermal-only generation provides a consistent and controlled framework for benchmarking and comparative analysis with existing literature. OPF formulation The OPF problem plays a fundamental role in the operation, planning, and secure control of modern power systems. The OPF formulation can be mathematically expressed as:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{OPF}} = {\mathrm{min}}\left( {F_{{{\mathrm{obj}}}} \left( {SV,CV} \right)} \right)$$\end{document} subject to:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g\left( {SV,CV} \right) = 0,lb \le h\left( {SV,CV} \right) \le ub$$\end{document} In this formulation, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{{{\mathrm{obj}}}}$$\end{document} denotes the objective function to be optimized, such as minimizing cost, or losses. The function \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g\left( {SV,CV} \right)$$\end{document} represents the equality constraints, typically power balance equations, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h\left( {SV,CV} \right)$$\end{document} defines the inequality constraints, bounded by lower and upper operational limits. The decision space of the OPF consists of two main components: state variables (SV) and control variables (CV)47, defined as follows:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$SV = \left[ {P_{{g1}} ,\,V_{{L1}} ,\, \ldots ,\,V_{{LN_{{LB}} }} ,\,Q_{{g1}} ,\, \ldots ,\,Q_{{g_{{N_{{GN}} }} }} ,\,SF_{1} ,\, \ldots ,\,SF_{{N_{{tl}} }} } \right]$$\end{document}4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$CV = \left[ {P_{g2} ,\, \ldots ,\,P_{{g_{{N_{GN} }} }} ,\,V_{g1} ,\, \ldots ,\,V_{{g_{{N_{GN} }} }} ,\,Q_{C1} ,\, \ldots ,\,Q_{{C_{{N_{C} }} }} ,\,T_{1} ,\, \ldots ,\,T_{{N_{TR} }} } \right]$$\end{document} where, the state variables typically include generator active (Pg1) at slack bus/reactive powers (Qg), load voltages (VL), and transmission line flows (SF), while the control variables consist of adjustable generation outputs (Pg), generator voltage setpoints (Vg), shunt compensator outputs (QC), and transformer tap ratios (T). These variables define the feasible operating region in which the OPF solution must be found. Also, the number of load busses, generators, lines, shunt compensators and tap transformers are symbolized by NLB, NGN, Ntl, NC and NTR. Objective function formulation Three objective functions are considered in this research48,49 as follows: Case study 1: Minimization of Quadratic Fuel Cost (QFC) Case study 1: Minimization of Quadratic Fuel Cost (QFC) Case study 1: Minimization of Quadratic Fuel Cost (QFC) One of the fundamental objectives in power system operation is to minimize the total fuel expenditure required to meet the system demand. Since thermal power units typically exhibit nonlinear fuel usage patterns, their costs are modeled using a quadratic relationship between generator output and fuel consumption where higher outputs require disproportionately more fuel.5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{min} \left( {SV,CV} \right) = \mathop \sum \limits_{i = 1}^{{N_{GN} }} A_{i} + B_{i} P_{{g_{i} }} + C_{i} P_{{g_{i} }}^{2} \left( {\$ /h} \right)\user2{ }$$\end{document} where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_{i} ,B_{i} ,C_{i}$$\end{document} represent cost coefficients of generator \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i.$$\end{document} b)Case study 2: Fuel Cost minimization with Valve-Point effects (FCVP) b)Case study 2: Fuel Cost minimization with Valve-Point effects (FCVP) Case study 2: Fuel Cost minimization with Valve-Point effects (FCVP) In many thermal units, valve-point loading introduces ripples in the fuel cost curve due to steam admission valves opening step-by-step. This creates a non-smooth, highly nonlinear cost function, making optimization more challenging as follows :6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{{min}} \left( {SV,CV} \right) = \mathop \sum \limits_{{i = 1}}^{{N_{{GN}} }} A_{i} + B_{i} P_{{g_{i} }} + C_{i} P_{{g_{i} }}^{2} + \left| {D_{i} \times sin\left( {E_{i} \left( {P_{{gi}}^{{min}} - P_{{gi}} } \right)} \right)} \right|~~\left( {\$ /h} \right)$$\end{document} where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D_{i} ,E_{i}$$\end{document} are the valve-point coefficients for generator \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i$$\end{document}; and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_{{g_{i} }}^{{{\mathrm{min}}}}$$\end{document} illustrates Minimum operating limit of generator \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i.$$\end{document} c)Case study 3: QFC Minimization with Multi-Fuel Options c)Case study 3: QFC Minimization with Multi-Fuel Options Case study 3: QFC Minimization with Multi-Fuel Options Some generating units are capable of burning different types of fuels, such as coal, gas, or oil, each having distinct cost characteristics. Power plants often switch fuels depending on price, availability, and emission policies. This creates a piecewise QFC as follows50:7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f\left( {P_{{g_{i} }} } \right) = \left\{ {\begin{array}{*{20}c} {A_{i,1} + B_{i,1} P_{{g_{i} }} + C_{i,1} P_{{g_{i} }}^{2} \left( {\$ /h} \right)} & {P_{{g_{i} ,1}} < P_{{g_{i} }} \le P_{{g_{i,2} }} } \\ . & . \\ {A_{i,m} + B_{i,m} P_{{g_{i} }} + C_{i,m} P_{{g_{i} }}^{2} \left( {\$ /h} \right)} & {P_{{g_{i,m} }} < P_{{g_{i} }} \le P_{{g_{i,m + 1} }} } \\ \end{array} } \right.$$\end{document} This formulation enables the OPF to choose the most economical fuel mix by capturing the operational and economic flexibility of multi-fuel units. where, m is the fuel segment index; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_{i,m} ,B_{i,m} ,C_{i,m}$$\end{document} denote the fuel cost coefficients for fuel type \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m$$\end{document} while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_{{g_{i,m} }}$$\end{document} is the power breakpoint between fuel segments. d)Case study 4: QFC Minimization with Prohibited Operating Zones (POZs) d)Case study 4: QFC Minimization with Prohibited Operating Zones (POZs) Case study 4: QFC Minimization with Prohibited Operating Zones (POZs) Mechanical vibration, worn-out bearings, problems with auxiliary equipment, or temperature restrictions can all make generators operate in areas that are not allowed. These prohibited zones partition the generator’s operating capability into disconnected intervals. The mathematic formulation for POZs is given by:8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_{{g_{i} }} \notin [P_{POZL} ,P_{POZU} ]_{m} ,m = 1,2, \ldots ,M_{i}$$\end{document} where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_{POZL,m}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_{POZU,m}$$\end{document} are the lower and upper limits of POZ interval \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m$$\end{document}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M_{i}$$\end{document} is the number of prohibited zones for generator \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i$$\end{document}. When POZs are considered, the optimization must avoid unsafe or unstable operating levels, enforcing both system reliability and equipment longevity while still aiming to minimize fuel cost. e)Case study 5: Real Power Losses (RPL) minimization e)Case study 5: Real Power Losses (RPL) minimization Case study 5: Real Power Losses (RPL) minimization Power losses in transmission networks are a natural consequence of current flow and resistance. Reducing these losses enhances voltage stability, lowers overall operating costs, and increases system efficiency and generation requirements.9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{min} \left( {SV,CV} \right) = \mathop \sum \limits_{i = 1}^{NB} P_{{g_{i} }} - \mathop \sum \limits_{i = 1}^{NB} P_{{d_{i} }} \user2{ }$$\end{document} where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$NB$$\end{document}: is the number of buses and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_{{d_{i} }}$$\end{document} is the load demand at bus \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i$$\end{document}(MW). To evaluate performance under different loading conditions, three scenarios are examined: Case 5a (Base Load): Loss minimization under normal operating conditions.Case 5b (Peak Load – 115%): Represents high-demand periods where losses increase significantly and maintaining stability becomes critical.Case 5c (Off-Peak Load – 85%): Reflects low-demand intervals where excess reactive power or overvoltage issues may arise. Case 5a (Base Load): Loss minimization under normal operating conditions. Case 5a (Base Load): Loss minimization under normal operating conditions. Case 5b (Peak Load – 115%): Represents high-demand periods where losses increase significantly and maintaining stability becomes critical. Case 5b (Peak Load – 115%): Represents high-demand periods where losses increase significantly and maintaining stability becomes critical. Case 5c (Off-Peak Load – 85%): Reflects low-demand intervals where excess reactive power or overvoltage issues may arise. Case 5c (Off-Peak Load – 85%): Reflects low-demand intervals where excess reactive power or overvoltage issues may arise. These cases collectively capture the electrical behavior of the grid under dynamic loading and allow assessing the optimizer’s robustness across varying system stress levels. Constraints The OPF problem must satisfy both equality constraints, which ensure electrical power balance, and inequality constraints, which enforce operational and security limits of power system components. These constraints guarantee that the obtained solution is not only optimal but also physically feasible and safe for practical operation. The equality constraints \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g\left( {SV,CV} \right)$$\end{document} represent the fundamental AC power flow equations, ensuring real and reactive power balance at the system level51. The inequality constraints \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h\left( {SV,CV} \right)$$\end{document} represent the operation and security limits of physical components and devices, which are described as follows:10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h \left(SV,CV\right)= \left\{\begin{array}{c}\begin{array}{c}{P}_{{g}_{j}}^{min}\le {P}_{{g}_{j}}\le {P}_{{g}_{j}}^{max} j =2:{N}_{GN}\\ \begin{array}{c}{V}_{{g}_{k}}^{min}\le {V}_{{g}_{k}}\le {V}_{{g}_{j}}^{max} j=1:{N}_{GN}\\ {Q}_{{g}_{j}}^{min}\le {Q}_{{g}_{j}}\le {Q}_{{g}_{j}}^{max} j=1:{N}_{GN}\\ {T}_{k}^{min} \le {T}_{k}\le {T}_{k}^{max} k=1:{T}_{{N}_{TR}}\end{array}\\ \begin{array}{c}{Q}_{{C}_{i}}^{min}\le {Q}_{{C}_{i}}\le {Q}_{{C}_{i}}^{max} i=1:{N}_{C}\\ { V}_{{L}_{m}}^{min}\le {V}_{{L}_{m}}\le {V}_{{L}_{m}}^{max} m=1:{N}_{LB}\\ {SF}_{n}\le {SF}_{n}^{max} n=1:{N}_{tl}\end{array}\end{array}\\ \end{array}\right.$$\end{document} This equation addresses the physical and operational impacts of the generator limits for the active power produced, voltage limits and reactive power constraints regarding overheating of generator field circuits. Also, transformer tap limits, shunt compensation bounds, load voltages restrictions, and line flow limits are handled. Constraint handling In OPF, some constraints, especially inequality constraints, may be violated during the search process of a population-based optimization algorithm. To ensure feasible solutions, a penalty-based augmented objective function (MT) is used as:11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M_{{\mathrm{T}}} = F_{{{\mathrm{min}}}} \left( {SV,CV} \right) + \lambda_{1} \mathop \sum \limits_{k = 1}^{{N_{GN} }} h\left( {P_{{g_{{1}} }} } \right) + \lambda_{2} \mathop \sum \limits_{j = 1}^{{N_{GN} }} h\left( {Q_{{g_{j} }} } \right) + \lambda_{3} \mathop \sum \limits_{m = 1}^{{N_{LB} }} h\left( {V_{{L_{m} }} } \right) + \lambda_{4} \mathop \sum \limits_{n = 1}^{{N_{tl} }} h\left( {SF_{{\mathrm{n}}} } \right)$$\end{document}12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h\left( {P_{{g_{{1}} }} } \right) = \left\{ {\begin{array}{*{20}c} {\left( {P_{{g_{{1}} }} - P_{{g_{{1}} }}^{max} } \right)^{2} } & {P_{{g_{{1}} }} > P_{{g_{{1}} }}^{max} } \\ {\left( {P_{{g_{{1}} }}^{min} - P_{{g_{{1}} }} } \right)^{2} } & {P_{{g_{{1}} }} < P_{{g_{{1}} }}^{min} } \\ \end{array} } \right.$$\end{document}13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h\left( {Q_{{g_{j} }} } \right) = \left\{ {\begin{array}{*{20}c} {\left( {Q_{{g_{j} }} - Q_{{g_{{\mathrm{j}}} }}^{max} } \right)^{2} } & {Q_{{g_{j} }} > Q_{{g_{{\mathrm{j}}} }}^{max} } \\ {\left( {Q_{{g_{{\mathrm{j}}} }}^{min} - Q_{{g_{j} }} } \right)^{2} } & {Q_{{g_{j} }} < Q_{{g_{{\mathrm{j}}} }}^{min} } \\ \end{array} } \right.$$\end{document}14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h\left( {V_{{L_{m} }} } \right) = \left\{ {\begin{array}{*{20}c} {\left( {V_{{L_{m} }} - V_{{L_{{\mathrm{m}}} }}^{max} } \right)^{2} } & {V_{{L_{m} }} > V_{{L_{{\mathrm{m}}} }}^{max} } \\ {\left( {V_{{L_{{\mathrm{m}}} }}^{min} - V_{{L_{m} }} } \right)^{2} } & {V_{{L_{m} }} < V_{{L_{{\mathrm{m}}} }}^{min} } \\ \end{array} } \right.$$\end{document}15\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h\left( {SF_{{\mathrm{n}}} } \right) = \left( {SF_{{\mathrm{n}}} - SF_{{\mathrm{n}}}^{max} } \right)^{2} SF_{{\mathrm{n}}} > SF_{{\mathrm{n}}}^{max}$$\end{document} where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M_{{\mathrm{T}}}$$\end{document} is the penalized fitness value; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda_{1} ,\lambda_{2} ,\lambda_{3} ,\lambda_{4}$$\end{document} are the penalty coefficients; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_{{g_{{1}} }}^{max}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_{{g_{{1}} }}^{min}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Q_{{g_{{\mathrm{j}}} }}^{max}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Q_{{g_{{\mathrm{j}}} }}^{min}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{{L_{{\mathrm{m}}} }}^{max}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{{L_{{\mathrm{m}}} }}^{min}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$SF_{{\mathrm{n}}}^{max}$$\end{document} are the maximum nad minimum real power output at slack generator, reactive power outputs at generator (j), voltage limits at load bus (m) and maximum line loading. Proposed MNRBO algorithm Standard NRBO algorithm The NRBO is a novel population-based metaheuristic algorithm that combines elements of Newton’s method with optimization challenges41. The main concept behind Newton’s method and optimization is rooted in their shared focus on finding solutions to complex problems by iteratively improving guesses towards an optimal result. Newton’s method, or Newton-Raphson method, is a gradient-based technique traditionally used to find the roots of a real-valued function. It starts with an initial guess for the solution. Using the function’s derivative to refine this guess iteratively52. It seeks to improve the accuracy by moving towards the root based on the slope of the tangent at the current estimate which can be mathematically formulated as follows:16\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{n + 1} = Z_{n} - \left( {{\raise0.7ex\hbox{${fit\left( {Z_{n} } \right)}$} \!\mathord{\left/ {\vphantom {{fit\left( {Z_{n} } \right)} {\left( {\frac{{\Delta fit\left( {Z_{n} } \right)}}{{\Delta Z_{n} }}} \right)}}}\right.\kern-0pt} \!\lower0.7ex\hbox{${\left( {\frac{{\Delta fit\left( {Z_{n} } \right)}}{{\Delta Z_{n} }}} \right)}$}}} \right)$$\end{document} This process is repeated until the guesses converge to the root. In optimization, the goal is to find the minimum (or maximum) of a function, often in a high-dimensional space. The challenge is similar to finding roots: identifying points where the function value is optimal. Optimization methods inspired by NRBO, adopt three principles. The first is gradient utilization which implements first and second derivatives to find roots, as Newton’s method, to guide the search for optimal solutions. These gradients help the algorithm navigate towards regions with better objective values. The second is iterative improvement which refines the solutions over iterations by adding randomness to maintain diversity in the population, allowing exploration and avoiding local optima. Finally, the balance between exploration and exploitation is maintained while a TAS is activated to avoid getting stuck in local optima. The NRBO follows a structured process that includes initialization, evaluation of fitness, application of Newton-Raphson-based search, and the use of operators to refine solutions. Initialization and fitness evaluation Each solution (or agent) in the population is represented by a decision vector (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z _{n}^{j}$$\end{document}) of dimension \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Dim$$\end{document}. Initial positions of the agents are randomly distributed within the search space:17\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z _{n}^{j} \left( {Iter = 0} \right) = Lower^{j} + \left( {\left( {Upper^{j} - Lower^{j} } \right) \times rand} \right){ }n = 1:N_{P} ,j = 1:Dim$$\end{document} where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Upper^{j}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Lower^{j}$$\end{document} are the boundaries of the decision variables; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$rand$$\end{document} is a random number between 0 and 1; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{P}$$\end{document} indicates the population size; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Iter$$\end{document} refers to the present iteration which is initially equals zero and it is increased gradually with one step till the maximum iterations (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Iter^{max}$$\end{document}). For each agent (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{n}$$\end{document}) in the population, the objective (fitness) function is evaluated as (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$fit\left( {Z_{n} } \right)$$\end{document}). After that, the agents are ranked based on their fitness values in order to identify the best agent (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z^{BEST}$$\end{document}) with the lowest fitness value (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{min}}\left( {fit\left( {Z_{n} } \right)} \right)$$\end{document}). and the worst agent (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z^{WORST}$$\end{document}) with the highest fitness value (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{max}}\left( {fit\left( {Z_{n} } \right)} \right)$$\end{document}). NRSR component Apply the Newton-Raphson search rule to update the position of each agent using gradient information. The NRSR is based on the following formula:18\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{n + 1} = Z_{n} - \left( {\frac{{fit\left( {Z_{n} + \Delta Z} \right) - fit\left( {Z_{n} - \Delta Z} \right)}}{{2 \times \left( {fit\left( {Z_{n} + \Delta Z} \right) + fit\left( {Z_{n} - \Delta Z} \right) - 2 \times fit\left( {Z_{n} } \right)} \right)}}} \right)$$\end{document} where19\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta Z = rand \times \left| {Z^{BEST} - Z_{n} } \right|$$\end{document} Therefore, certain adjustments are required to control the population-based search effectively defining the NRSR as follows:20\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$NRSR\left( {Z_{1} ,Z_{2} } \right) = randn \times \left( {\frac{{\left( {Z_{1} - Z_{2} } \right) \times \Delta Z_{12} }}{{2 \times \left( {Z_{1} + Z_{2} - 2 \times Z_{n} } \right)}}} \right)$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$randn$$\end{document} represents a randomized number following the normal distribution; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{2}$$\end{document} are two specified solution vectors while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{n}$$\end{document} is the current solution and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta Z_{12}$$\end{document} indicates its weighted differences as described as follows: 21\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta Z_{12} = rand \times \left| {Z_{1} - Z_{2} } \right|$$\end{document} Using the suggest NRSR, a key advantage of this approach is that it updates based on the position (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{n}$$\end{document}) rather than its fitness ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$fit\left( {Z_{n} } \right)$$\end{document}), which improves computational efficiency. New solutions are generated by combining three components with random perturbations as follows: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z _{n}^{j} \left( {Iter + 1} \right) = Rd_{1} \times \left( {Rd_{1} \times Z1 _{n}^{j} \left( {Iter} \right) + \left( {1 - Rd_{1} } \right) \times Z2 _{n}^{j} \left( {Iter} \right)} \right){ + }\left( {\left( {1 - Rd_{1} } \right) \times Z3 _{n}^{j} \left( {Iter} \right)} \right){ }$$\end{document} 22\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${ }n = 1:N_{P} ,j = 1:Dim$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Rd_{1}$$\end{document} represents a randomized number in the range (0,1); \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z1\left( {Iter} \right)$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z2\left( {Iter} \right)$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z3\left( {Iter} \right)_{2}$$\end{document} are three iterative components which are mathematically defined as: 23\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z1_{n} \left( {Iter} \right) = Z_{n} \left( {Iter} \right) - NRSR\left( {Y_{W} ,Y_{V} } \right) + \rho$$\end{document} 24\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z2_{n} \left( {Iter} \right) = Z^{BEST} - NRSR\left( {Y_{W} ,Y_{V} } \right) + \rho$$\end{document} 25\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho = a \times \left( {Z^{BEST} - Z_{n} \left( {Iter} \right)} \right) + b \times \left( {Z_{R1} \left( {Iter} \right) - Z_{R2} \left( {Iter} \right)} \right)$$\end{document} 26\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Y_{W} = Rd_{2} \times \left( {\left( {Mean\left( {Z_{n} - randn \times NRSR\left( {Z^{WORST} ,Z^{BEST} } \right),Z_{n} } \right)} \right) + Rd_{2} \times \Delta Z} \right)$$\end{document} 27\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Y_{V} = Rd_{2} \times \left( {\left( {Mean\left( {Z_{n} - randn \times NRSR\left( {Z^{WORST} ,Z^{BEST} } \right),Z_{n} } \right)} \right) - Rd_{2} \times \Delta Z} \right)$$\end{document} 28\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z3_{n} \left( {Iter} \right) = Z_{n} \left( {Iter} \right) - \delta \left( {Iter} \right) + a \times \left( {Z2_{n} \left( {Iter} \right) - Z1_{n} \left( {Iter} \right)} \right)$$\end{document} 29\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta \left( {Iter} \right) = \left( {1 - \left( {\frac{2 \times Iter}{{Iter^{max} }}} \right)} \right)^{5}$$\end{document} where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{R1} \left( {Iter} \right)$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{R2} \left( {Iter} \right)$$\end{document} are two distinct randomly selected solutions from the population while a and b are random numbers between 0 and 1. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Rd_{2}$$\end{document} represents a randomized number in the range (0,1). TAS component In the NRBO, the TAS is suggested to escape local optima. TAS generates new positions based on the best agent (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z^{BEST}$$\end{document}) and the current agent (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{n} \left( {Iter} \right)$$\end{document}) as follows: 30\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{TAO} \left( {Iter} \right) = \left\{ {\begin{array}{*{20}c} {Z_{n} \left( {Iter + 1} \right) + XA\left( {Iter} \right) + XB\left( {Iter} \right)} & {if \mu_{a} < 0.5 } \\ {Z^{BEST} + XA\left( {Iter} \right) + XB\left( {Iter} \right)} & {Otherwise} \\ \end{array} } \right.$$\end{document} 31\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$XA\left( {Iter} \right) = \theta _{1} \times \left( {\left( {\mu_{a} \times Z^{BEST} } \right) - \left( {\mu_{b} \times Z_{n} \left( {Iter} \right)} \right)} \right)$$\end{document} 32\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$XB\left( {Iter} \right) = \theta _{2} \times \delta \left( {Iter} \right) \times \left( {\left( {\mu_{a} \times Mean\left( {Z_{n} } \right)} \right) - \left( {\mu_{b} \times Z_{n} \left( {Iter} \right)} \right)} \right)$$\end{document} 33\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu_{a} = \left( {1 - \beta } \right) + 3 \times \beta \times rand$$\end{document} 34\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu_{b} = \left( {1 - \beta } \right) + \beta \times rand$$\end{document} 35\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta = \left\{ {\begin{array}{*{20}c} 1 & {if Rd_{3} > 0.5 } \\ 0 & {Otherwise} \\ \end{array} } \right.$$\end{document} The randomness in selecting the parameters \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu_{a}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu_{b}$$\end{document}increases the diversity of the population, allowing it to escape local optima and enhancing overall diversification. Figure 1 displays the main steps of the presented NRBO. Fig. 1NRBO steps. NRBO steps. Proposed MNRBO version The NRBO search mechanism, primarily governed by NRSR and TAS, generates new candidate solutions independently, reducing population diversity and potentially leading to premature stagnation or local optima. The proposed MNRBO can be viewed as a hybrid between Newtonian-based local search and adaptive evolutionary population control, addressing a common weakness in NRBO of the lack of diversity regulation. First, the ACM in MNRBO uses a crossover probability (ϕ) that changes automatically during the iterations. This probability decreases linearly from a maximum (ϕMax) to a minimum (ϕMin) value as the iterations progress, as described in Eq. (29). At the beginning, when ϕ is high, frequent crossover operations occur, allowing broad exploration of the search space through extensive information exchange between individuals. This helps the algorithm discover promising regions efficiently. As the MNRBO advances, ϕ gradually decreases, limiting random perturbations and allowing the search to concentrate on fine-tuning the best solutions.36\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi \left( {Iter} \right) = \phi^{Min} - \left( {\left( {\phi^{Max} - \phi^{Min} } \right) \times \left( {1 - \frac{Iter}{{Iter^{max} }}} \right)} \right)$$\end{document} The crossover operation can be activated to each control variable as:37\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{n,j} \left( {Iter + 1} \right) = \left\{ {\begin{array}{*{20}c} {Z_{n,j} \left( {Iter} \right)} & {if r_{j} < \phi \left( {Iter} \right)} \\ {Z_{n,j} \left( {Iter + 1} \right)} & {Otherwise} \\ \end{array} } \right.$$\end{document} where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{n,j} \left( {Iter} \right)$$\end{document} is the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$j^{th}$$\end{document} dimensional part of the present n^th search solution, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r_{j} \in$$\end{document} (0,1) constitutes a randomly generated amount, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{n,j} \left( {Iter + 1} \right)$$\end{document} is the freshly created value derived from the Newtonian and TAS operations. This adaptive crossover strategy functions similarly to the recombination process in DE but with a self-regulating behavior driven by ϕ. As a result, MNRBO becomes a dynamic and self-evolving optimizer that naturally shifts its focus from exploration to exploitation. This capability enhances its stability, robustness, and effectiveness in solving complex, multimodal, and high-dimensional optimization problems. In the proposed MNRBO, instead of Eq. (29), the Sigmoid decay mode is introduced to dynamically regulate the convergence progress during iterations. It defines the control parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta \left( {{\mathrm{Iter}}} \right)$$\end{document}as:38\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta \left( {Iter} \right) = \frac{\delta o}{{1 + exp\left( {\frac{\alpha }{{Iter^{max} }} \times \left( {Iter - \left( {\frac{{Iter^{max} }}{2}} \right)} \right)} \right)}}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta_{0}$$\end{document} is the initial maximum value (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta_{0}$$\end{document} = 1), α controls the steepness of the curve (α = 10), and Itermax is the maximum number of iterations. Figure 2 illustrates the variation of the control parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta \left( {{\mathrm{Iter}}} \right)$$\end{document} across iterations for both the polynomial decay function (Eq. 29), in the upper plot, and the Sigmoid decay mode (Eq. 38) displayed in the lower plot. In the polynomial decay curve, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta$$\end{document} exhibits high perturbation magnitudes at both the beginning and the end of the search process, tending toward values close to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pm$$\end{document} 1. Fig. 2Polynomial versus Sigmoid decay modes. Polynomial versus Sigmoid decay modes. This behavior implies that large search steps may occur not only during the early exploration phase but also near the final iterations. As a result, the NRBO algorithm can experience instability or oscillatory movements near convergence, reducing its precision in local refinement. In contrast, the Sigmoid decay mode produces a smooth, monotonic reduction in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta$$\end{document} as iterations progress. The value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta$$\end{document} decreases gradually from high exploration levels to low perturbation magnitudes, concentrating the MNRBO’s adjustments within a controlled range during later stages. This adaptive transition allows the search to begin with wide, diverse exploration of the solution space and progressively shift toward focused local exploitation. Consequently, the sigmoid-based decay enhances convergence stability, prevents unnecessary large movements near the optimum, and provides a more reliable balance between exploration and exploitation. Figure 3 displays the main steps of the MNRBO including the ACM and the Sigmoid decay mode. Also, the detailed steps is depicted in Algorithm 1. The ACM allows the algorithm to share directional knowledge among agents. Also, the Sigmoid Decay Mode provides a nonlinear time-evolution control, offering a smoother transition between exploration and exploitation phases. These additions turn NRBO into a self-evolving population framework rather than a fixed-behavior searcher. However, these modifications come at a computational cost. Each iteration requires extra vector recombination and possible re-evaluation of fitness after crossover. The adaptive parameters (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi \left( {Iter} \right)$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta \left( {Iter} \right)$$\end{document}) require tuning to avoid under- or over-exploitation. Despite the cost, the enhancements don’t merely “add randomness,” but introduce adaptive behavioral learning to the Newtonian mechanism. Fig. 3MNRBO steps. MNRBO steps. Algorithm 1 Proposed MNRBO. Algorithm 1 Proposed MNRBO. Application and results The proposed MNRBO is evaluated on the standard IEEE 30-bus transmission network and its performance is compared against the original NRBO. As shown in Fig. 4, the IEEE 30-bus system consists of 30 buses, 41 transmission lines, six generators, nine Shunt Compensator banks, and four on-load tap-changing transformers. The generator voltage limits range from 0.9500 p.u. to 1.1000 p.u., and the system’s total generation capacity is 900.2 MW. Table 1 summarizes the main structural characteristics of the system. All system parameters, including line data, generator cost coefficients, emission coefficients, and prohibited operating zones (POZs), follow the standard data sets documented in50. Table 1IEEE 30-bus system: Main characteristics.System characteristicsDetailsBuses3050Branches4150Control variables24Generators6 Buses: 1, 2, 5, 8, 11 and 13Shunts9 Buses: 10, 12, 15, 17, 20, 21,23, 24 and 29Transformers4 Branches: 11, 12, 15 and 36 Table 1IEEE 30-bus system: Main characteristics.System characteristicsDetailsBuses3050Branches4150Control variables24Generators6 Buses: 1, 2, 5, 8, 11 and 13Shunts9 Buses: 10, 12, 15, 17, 20, 21,23, 24 and 29Transformers4 Branches: 11, 12, 15 and 36 IEEE 30-bus system: Main characteristics. A total of seven OPF case studies are conducted in this work, beginning with basic QFC minimization in Case 1 and progressively adding complexities such as valve-point effects (Case 2), multi-fuel characteristics (Case 3), POZs (Case 4), and loss minimization (Case 5a, b and c) under varying load conditions. The proposed MNRBO and the original NRBO are evaluated while three other recent algorithms are also implemented of Somersaulting Spider Optimizer (SSO), 202553, Birds of prey-based optimization (BPBO), 202554, , and Pied kingfisher optimizer: a (PKO), 202455, . To ensure a fair and unbiased comparative evaluation, all competing algorithms were implemented under identical experimental conditions, including the same population size, number of iterations, and consequently the same number of objective function evaluations. Each OPF scenario is executed 20 independent simulation runs using a population size of 50 and 200 iterations to ensure consistent convergence behavior and statistical reliability of the results. Based on that settings, the number of function of evaluations is 10,000 for all applied algorithms (NRBO, MNRBO, BPBO and SSO) except the PKO which has 20,000. This is due that NRBO, MNRBO, BPBO and SSO has only function evaluation for each search solution in each iteration while PKO has two times. Moreover, no algorithm-specific parameter tuning was performed in favor of the proposed method; instead, all parameters were selected based on commonly recommended values in the literature. Fig. 4IEEE 30 bus system. IEEE 30 bus system. Case study 1: QFC minimization Table 2 lists the optimal values of all OPF control variables obtained by the five competing algorithms: BPBO, NRBO, MNRBO, PKO, and SSO. These variables include generator active-power outputs, generator voltage magnitudes, transformer tap settings, and Shunt Compensator injections. As shown, the MNRBO produces the most economically feasible solution, with the smallest QFC of 799.0837 $/h while BPBO, NRBO, PKO and SSO achieve QFC values of 799.2847, 799.5722, 799.2161 and 799.3725, respectively. In this regard, Fig. 5 displays (a) the best and (b) the average convergence trajectories of all algorithms. As shown in Fig. 5(a), the MNRBO converges much faster, achieving a steep descent in the early iterations and stabilizing smoothly near the optimum. From Fig. 5(b), the MNRBO average curve stays significantly lower at the second half of iterations, confirming high repeatability and low variance across runs. The average curves, shown in Fig. 5(b), validate that MNRBO is not only the best performer but also the most reliable across multiple executions. To further verify the feasibility and effectiveness of the applied algorithms, all obtained solutions were carefully examined with respect to the operational constraints of the OPF problem, including generator active/reactive power limits, bus voltage limits, transformer tap-setting bounds, shunt compensator constraints, and transmission line operating limits. As shown in Table 2, the feasibility rate is always 100% for all acquired operating points that confirmed that all control and state variables remained within their allowable operating boundaries for all investigated case studies and independent runs. The overall penalty values associated with the acquired optimal solutions were found to be equal to zero, indicating that no constraint violations occurred during the optimization process. This demonstrates the capability of the proposed MNRBO algorithm to efficiently satisfy system constraints while maintaining high-quality optimal solutions and stable convergence behavior. Table 2Optimum values of control variables for case 1.S. No.Control variableBPBONRBOProposed MNRBOPKOSSOActive power output of the generator (MW) 1Pg248.6649947.8783548.6047449.117447.64719 2Pg521.3375422.0786621.3766321.2272121.22314 3Pg820.8171922.842821.2187820.527121.13092 4Pg1112.4813610.9227811.8783712.0416211.73226 5Pg1312.002811212.000112.2049412Generator voltage 6Vg11.0999961.11.0999971.099991.1 7Vg21.0866861.0888891.087841.0853451.085221 8Vg51.0612561.0624451.0618861.0582921.058852 9Vg81.0681731.0702381.0698091.0684021.066941 10Vg111.0999871.11.11.0985821.098196 11Vg131.099731.11.0999971.0999171.098358Tap-setting of transformer 12T6−91.0505490.9984411.0588160.9972920.962906 13T6−100.9004430.9885780.900010.9947551.036989 14T4−121.0162550.9874890.9912711.0094121.029387 15T28−270.9716480.9769920.9663050.975380.980349Shunt compensator 16Qc101.4513582.2566824.9998543.1174312.864846 17Qc124.9025450.7581594.9997843.8105210.872138 18Qc153.0535843.0193464.9999754.4841493.886923 19Qc172.4931733.9760514.9999684.4553144.159654 20Qc202.0181093.4137894.5848132.5533071.879272 21Qc214.9996252.57330154.4916924.541805 22Qc232.879232.7511372.8298213.1262411.393038 23Qc241.9251761.0175354.9999954.9841043.638678 24Qc292.1690671.5984972.2961132.1619484.057894QFC ($/h)799.2847799.5722799.0837799.2161799.3725Feasibilty rate100%100%100%100%100% Table 2Optimum values of control variables for case 1.S. No.Control variableBPBONRBOProposed MNRBOPKOSSOActive power output of the generator (MW) 1Pg248.6649947.8783548.6047449.117447.64719 2Pg521.3375422.0786621.3766321.2272121.22314 3Pg820.8171922.842821.2187820.527121.13092 4Pg1112.4813610.9227811.8783712.0416211.73226 5Pg1312.002811212.000112.2049412Generator voltage 6Vg11.0999961.11.0999971.099991.1 7Vg21.0866861.0888891.087841.0853451.085221 8Vg51.0612561.0624451.0618861.0582921.058852 9Vg81.0681731.0702381.0698091.0684021.066941 10Vg111.0999871.11.11.0985821.098196 11Vg131.099731.11.0999971.0999171.098358Tap-setting of transformer 12T6−91.0505490.9984411.0588160.9972920.962906 13T6−100.9004430.9885780.900010.9947551.036989 14T4−121.0162550.9874890.9912711.0094121.029387 15T28−270.9716480.9769920.9663050.975380.980349Shunt compensator 16Qc101.4513582.2566824.9998543.1174312.864846 17Qc124.9025450.7581594.9997843.8105210.872138 18Qc153.0535843.0193464.9999754.4841493.886923 19Qc172.4931733.9760514.9999684.4553144.159654 20Qc202.0181093.4137894.5848132.5533071.879272 21Qc214.9996252.57330154.4916924.541805 22Qc232.879232.7511372.8298213.1262411.393038 23Qc241.9251761.0175354.9999954.9841043.638678 24Qc292.1690671.5984972.2961132.1619484.057894QFC ($/h)799.2847799.5722799.0837799.2161799.3725Feasibilty rate100%100%100%100%100% Optimum values of control variables for case 1. Fig. 5Convergence of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 1. (a) Best convergences, (b) average convergences. Convergence of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 1. (a) Best convergences, (b) average convergences. Additionally, Fig. 6 shows the boxplots of QFC values for Case 1 presenting the statistical distribution of the obtained costs (20 runs) for each algorithm. As shown, the proposed MNRBO shows the smallest box width, and tight whiskers, demonstrating extremely consistent convergence behavior. NRBO shows the widest box and highest spread, confirming its sensitivity to search conditions and inconsistent numerical stability. BPBO and PKO show moderate variation with occasional higher outliers. SSO remains relatively stable but performs worse than MNRBO in terms of best and mean values. Table 3 reports the QFC results from 20 independent simulation runs for every algorithm. The proposed MNRBO consistently yields costs within a very narrow range (799.0837–799.3101), showing excellent repeatability. In contrast, NRBO displays highly varied outcomes between 799.5722 and 814.6369, indicating significant convergence instability. Both BPBO and PKO show acceptable but higher variability compared to MNRBO, while SSO, despite its stability, fails to achieve the same optimal performance as MNRBO. Table 4 further summarizes the statistical metrics, including the minimum, mean, maximum, and standard deviation of the objective values across the 20 runs. Fig. 6Boxplots of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 1. Boxplots of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 1. Table 3Obtained QFC objectives of the twenty different runs of MNRBO, NRBO, BPBO, PKO and SSO for Case 1.Run No.BPBONRBOProposed MNRBOPKOSSO1800.7606801.2148799.0855799.6121799.41832799.3709801.7903799.1471799.2161799.67293799.4593805.7411799.1219799.3653799.47664799.2847801.8387799.1697799.3075799.56635799.4448803.7155799.3101799.2681799.59116799.4624803.4959799.1334799.5381799.40087799.5112804.1627799.1722799.7454799.52998799.546799.5722799.1825799.4471799.43399799.5359809.686799.0919799.2693799.677110799.4253800.4608799.0837799.3235799.454611799.462800.7138799.1045799.2361799.397612799.6005800.1095799.1245799.353799.530113799.4382801.8014799.1629799.637799.47814799.7974801.9323799.0843799.4046799.697615799.7087800.6193799.1853799.7323799.541816800.0711801.1422799.1695799.8601799.497117799.6515800.0534799.132799.4432799.613318800.1504814.6369799.1055799.824799.573919799.5987802.5005799.1058800.1228799.372520799.8484808.0539799.1861799.3764799.5915 Table 3Obtained QFC objectives of the twenty different runs of MNRBO, NRBO, BPBO, PKO and SSO for Case 1.Run No.BPBONRBOProposed MNRBOPKOSSO1800.7606801.2148799.0855799.6121799.41832799.3709801.7903799.1471799.2161799.67293799.4593805.7411799.1219799.3653799.47664799.2847801.8387799.1697799.3075799.56635799.4448803.7155799.3101799.2681799.59116799.4624803.4959799.1334799.5381799.40087799.5112804.1627799.1722799.7454799.52998799.546799.5722799.1825799.4471799.43399799.5359809.686799.0919799.2693799.677110799.4253800.4608799.0837799.3235799.454611799.462800.7138799.1045799.2361799.397612799.6005800.1095799.1245799.353799.530113799.4382801.8014799.1629799.637799.47814799.7974801.9323799.0843799.4046799.697615799.7087800.6193799.1853799.7323799.541816800.0711801.1422799.1695799.8601799.497117799.6515800.0534799.132799.4432799.613318800.1504814.6369799.1055799.824799.573919799.5987802.5005799.1058800.1228799.372520799.8484808.0539799.1861799.3764799.5915 Obtained QFC objectives of the twenty different runs of MNRBO, NRBO, BPBO, PKO and SSO for Case 1. Table 4Statistical indicators of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 1.Case 1BPBONRBOProposed MNRBOPKOSSOMin QFC799.2847799.5722799.0837799.2161799.3725Average QFC799.6564803.1621799.1429799.5041799.5257Max QFC800.7606814.6369799.3101800.1228799.6976STD0.3424873.7977880.0532820.2475940.09751 Table 4Statistical indicators of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 1.Case 1BPBONRBOProposed MNRBOPKOSSOMin QFC799.2847799.5722799.0837799.2161799.3725Average QFC799.6564803.1621799.1429799.5041799.5257Max QFC800.7606814.6369799.3101800.1228799.6976STD0.3424873.7977880.0532820.2475940.09751 Statistical indicators of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 1. As shown, MNRBO’s superior performance is confirmed by the lowest minimum, mean, and maximum expenses. As can be seen, MNRBO has an exceptionally low standard deviation of 0.053, which is a reduction of 98.6% when compared to NRBO, 84.5% when compared to BPBO, 78.5% when compared to PKO, and 45.4% when compared to SSO. This dramatic decrease demonstrates MNRBO’s resilience to random variations. On the other hand, NRBO performs poorly statistically, with a large cost spread and a high standard deviation (3.798), suggesting uneven convergence in the absence of adaptive mechanisms. The performance of BPBO, PKO, and SSO is moderate, but none of them can equal the accuracy and stability of MNRBO, which stands out among the examined algorithms for its greater dependability and cost effectiveness. For QFC minimization, Table 5 presents a comprehensive comparison results obtained by the proposed MNRBO and several reported methods. The suggested MNRBO outperforms all rivals in terms of solution quality with the lowest operating cost of 799.0837 $/h. The MNRBO shows even greater improvements when compared to other reported algorithms in the OPF literature, including DE, enhanced self-adaptive DE, improved moth-flame optimisation, moth swarm optimisation, and the modified grasshopper optimiser. When compared to these approaches, MNRBO’s cost reduction varies from 0.16% to 0.18%, which translates into savings of 1.30–1.43 $/h. These results clearly reaffirm its superior search capability and robustness in quadratic cost OPF environments. The obtained results confirm that the proposed MNRBO provides a more effective exploration–exploitation balance, enabling it to locate higher-quality operating points while avoiding premature convergence. Consequently, MNRBO demonstrates superior optimization capability in terms of robustness, and solution accuracy for QFC-OPF problems, establishing its competitiveness against both recent and well-established optimization techniques. Table 5Assessment of OPF results obtained By MNRBO and other reported algorithms For QFC minimization.AlgorithmQFC ($/h)Proposed MNRBO799.0837PKO799.2161BPBO799.2847SSO799.3725NRBO799.5722DE [56]800.4148Improved moth-flame optimization57800.3848Enhanced self-adaptive DE58800.5141Modified grasshopper optimizer59800.4744Moth swarm algorithm60800.5099 Table 5Assessment of OPF results obtained By MNRBO and other reported algorithms For QFC minimization.AlgorithmQFC ($/h)Proposed MNRBO799.0837PKO799.2161BPBO799.2847SSO799.3725NRBO799.5722DE [56]800.4148Improved moth-flame optimization57800.3848Enhanced self-adaptive DE58800.5141Modified grasshopper optimizer59800.4744Moth swarm algorithm60800.5099 Assessment of OPF results obtained By MNRBO and other reported algorithms For QFC minimization. Case study 2: Fuel Cost minimization with Valve-Point effects (FCVP) This case evaluates the ability of the proposed MNRBO to navigate the search space compared to BPBO, NRBO, PKO, and SSO where FCVP effects make the OPF problem significantly more difficult by introducing highly non-smooth ripples into the fuel-cost function. Table 6 lists the OPF control settings obtained by each applied algorithm. As shown, the MNRBO achieves the lowest fuel cost of 832.9199 $/h, outperforming all competitors in terms of solution quality. It presents hourly cost reductions by 3.164 $/h, 0.465 $/h, 0.494 $/h, and 0.763 $/h against NRBO, BPBO, PKO, and SSO which attained costs of 836.084, 833.385, 833.413 and 833.683, respectively. Even small cost differences in thermal OPF represent significant savings in long-term operation. The plots in Fig. 7 illustrate how quickly and smoothly each algorithm reduces the FCVP objective over iterations. As shown, the MNRBO converges the fastest (after approximately 40% of the iterations), reaching a stable minimum early with minimal oscillations. Table 6Optimum values of control variables for case 2.S. No.Control variableBPBONRBOProposed MNRBOPKOSSOActive power output of the generator (MW) 1Pg247.7589348.3034147.6890446.9759848.25367 2Pg519.2568916.1638319.0403519.8837318.60368 3Pg810.1075312.1085410.012331010.56448 4Pg1110.00194101010.0746810 5Pg1312.0159612.97961121212.09022Generator voltage 6Vg11.099881.0999991.11.0999991.1 7Vg21.0805361.0807371.0802141.0788251.08033 8Vg51.0525261.0586741.0556531.0566891.049145 9Vg81.0634651.074011.0656441.0688361.059486 10Vg111.0893211.0754111.11.0885621.073493 11Vg131.0706611.0999991.11.0999941.083295Tap-setting of transformer 12T6−91.070241.0854680.9830531.0297750.990358 13T6−100.9622581.0324981.0999991.0034520.936922 14T4−121.0604490.9950991.0808681.0970041.017327 15T28−271.0057491.0814980.9930441.0016561.00203Shunt compensator 16Qc100.0476440.5267054.9979914.1549732.825224 17Qc120.2519192.48100154.7239882.626659 18Qc154.8198192.3395923.5043930.1833414.320549 19Qc174.9948370.3342794.2702291.0495623.509396 20Qc201.8169384.8078284.9999944.061552.137293 21Qc211.2629741.5186563.7030044.9938491.220714 22Qc234.7432320.6068044.7784152.945211.088067 23Qc240.9760754.73196850.6827572.454902 24Qc294.8858981.3191071.3452311.2144153.452557QFC ($/h)833.3847836.084832.9199833.4134833.6827 Table 6Optimum values of control variables for case 2.S. No.Control variableBPBONRBOProposed MNRBOPKOSSOActive power output of the generator (MW) 1Pg247.7589348.3034147.6890446.9759848.25367 2Pg519.2568916.1638319.0403519.8837318.60368 3Pg810.1075312.1085410.012331010.56448 4Pg1110.00194101010.0746810 5Pg1312.0159612.97961121212.09022Generator voltage 6Vg11.099881.0999991.11.0999991.1 7Vg21.0805361.0807371.0802141.0788251.08033 8Vg51.0525261.0586741.0556531.0566891.049145 9Vg81.0634651.074011.0656441.0688361.059486 10Vg111.0893211.0754111.11.0885621.073493 11Vg131.0706611.0999991.11.0999941.083295Tap-setting of transformer 12T6−91.070241.0854680.9830531.0297750.990358 13T6−100.9622581.0324981.0999991.0034520.936922 14T4−121.0604490.9950991.0808681.0970041.017327 15T28−271.0057491.0814980.9930441.0016561.00203Shunt compensator 16Qc100.0476440.5267054.9979914.1549732.825224 17Qc120.2519192.48100154.7239882.626659 18Qc154.8198192.3395923.5043930.1833414.320549 19Qc174.9948370.3342794.2702291.0495623.509396 20Qc201.8169384.8078284.9999944.061552.137293 21Qc211.2629741.5186563.7030044.9938491.220714 22Qc234.7432320.6068044.7784152.945211.088067 23Qc240.9760754.73196850.6827572.454902 24Qc294.8858981.3191071.3452311.2144153.452557QFC ($/h)833.3847836.084832.9199833.4134833.6827 Optimum values of control variables for case 2. Fig. 7Convergence of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 2. (a) Best convergences, (b) average convergences. Convergence of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 2. (a) Best convergences, (b) average convergences. Additionally, Fig. 8 shows the boxplots of QFC values for Case 2 while Table 7 displays the related statistical indicators of MNRBO, NRBO, BPBO, PKO and SSO algorithms. As shown, the MNRBO has the narrowest box and tightest whiskers, indicating high consistency with the smallest standard deviation of 0.224. NRBO shows extremely wide variation, confirming its sensitivity to rugged FCVP search space. BPBO and PKO show moderate spreads, with occasional high-cost outliers. SSO is generally stable but consistently inferior to MNRBO in terms of the best and mean costs. The proposed MNRBO declares significant quantitative improvement in stability in terms of the standard deviation with 93.3%, 86.7%, 66.7%, and 71.4% reduction compared to NRBO, BPBO, PKO, and SSO respectively which attain counterparts of 3.326, 1.684, 0.673, and 0.782. This verifies that MNRBO is up to 14 times more stable than NRBO and 4–7 times more stable than SSO, PKO, or BPBO. Moreover, MNRBO is the only algorithm with no large outliers, showing reliability under repeated simulations. The proposed MNRBO consistently stays within 833.0 ± 0.3, showing excellent robustness while NRBO fluctuates wildly from 836 to 847, proving instability; BPBO and PKO have moderate but noticeable variability; and SSO remains stable but never outperforms MNRBO in any run. The proposed MNRBO demonstrates a substantial superiority over the standard NRBO. In quantitative terms, MNRBO achieves a reduction in mean operational cost from 841.126 $/h to 833.148 $/h, reflecting an improvement of 0.95%. Additionally, the maximum observed cost is decreased by 13.647 $/h, translating to a 1.61% reduction. These results serve to underline the effectiveness of MNRBO in circumventing inferior-quality solutions. Moreover, the standard deviation of the costs significantly drops from 3.326 to 0.224, illustrating an impressive enhancement in stability estimated at 93.3%. Such stability improvements are attributed to MNRBO’s adaptive mechanisms, which markedly enhance its consistency and robustness. This is particularly evident in its handling of the highly non-smooth cost features caused by valve-point effects, which typically present significant challenges in optimization scenarios. Fig. 8Boxplots of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 2. Boxplots of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 2. Table 7Statistical indicators of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 2.Case 2BPBONRBOProposedMNRBOPKOSSOMin QFC833.3847836.084832.9199833.4134833.6827Average QFC835.1768841.1258833.1479834.3284834.7247Max QFC839.5562847.5841833.9373835.4919836.6173STD1.6845183.3264180.2240540.6732980.782385 Table 7Statistical indicators of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 2.Case 2BPBONRBOProposedMNRBOPKOSSOMin QFC833.3847836.084832.9199833.4134833.6827Average QFC835.1768841.1258833.1479834.3284834.7247Max QFC839.5562847.5841833.9373835.4919836.6173STD1.6845183.3264180.2240540.6732980.782385 Statistical indicators of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 2. For this case, the one-way ANOVA analysis is implemented and their results is illustrated in Table 8. This analysis demonstrates that the obtained optimization performances differ significantly among the investigated algorithms. Since the p-value is substantially lower than the significance threshold of 0.05, the null hypothesis of equal mean performances is rejected. Additionally, the Tukey–Kramer post-hoc analysis is shown in Table 9 which further clarifies the pairwise statistical relationships among the compared methods. The results reveal that the original NRBO algorithm performs significantly worse than all other algorithms, including BPBO, MNRBO, PKO, and SSO. This is confirmed by extremely small p-values ranging from 10–17 to 10–25, together with confidence intervals that do not cross zero. The proposed MNRBO demonstrates substantial superiority over both NRBO and SSO. In particular, the comparison between MNRBO and NRBO shows a highly significant improvement with a mean difference of 7.978 and a p-value of 3.1894 × 10–25. Moreover, MNRBO also statistically significantly better performance compared to SSO with a p-value of 0.0389, indicating that the proposed modifications significantly improve optimization stability and convergence quality. Additionally, the comparison between BPBO and MNRBO indicates a statistically significant difference (p = 0.003242), demonstrating that MNRBO achieves superior optimization performance compared to BPBO. However, the differences between MNRBO and PKO are not statistically significant, indicating that both algorithms provide relatively comparable performance under the tested operating conditions. Furthermore, PKO and SSO exhibit statistically similar behavior, as reflected by the high p-value of 0.9507, suggesting that their mean performances are nearly equivalent. Table 8One-way anova results for Case 2.Statistical testValueTest typeOne-Way ANOVANumber of algorithms5Significance level0.05P-value0.0000000000Statistical decisionSignificant differences existamong algorithms Table 8One-way anova results for Case 2.Statistical testValueTest typeOne-Way ANOVANumber of algorithms5Significance level0.05P-value0.0000000000Statistical decisionSignificant differences existamong algorithms One-way anova results for Case 2. Significant differences exist among algorithms Table 9Tukey–kramer post-hoc multiple comparison results for Case 2.ComparisonMean differenceLower CIUpper CIP-valueBPBO vs. NRBO-5.9490-7.4735-4.42452.0324 × 10–17BPBO vs. MNRBO2.02900.50453.55340.003242BPBO vs. PKO0.8485-0.67602.37290.5344BPBO vs. SSO0.4521-1.07241.97660.9224NRBO vs. MNRBO7.97806.45359.50243.1894 × 10–25NRBO vs. PKO6.79755.27308.32199.8382 × 10–21NRBO vs. SSO6.40114.87667.92563.4117 × 10–19MNRBO vs. PKO-1.1805-2.70500.34400.2066MNRBO vs. SSO-1.5769-3.1013-0.05240.038916PKO vs. SSO-0.3964-1.92081.12810.9507 Table 9Tukey–kramer post-hoc multiple comparison results for Case 2.ComparisonMean differenceLower CIUpper CIP-valueBPBO vs. NRBO-5.9490-7.4735-4.42452.0324 × 10–17BPBO vs. MNRBO2.02900.50453.55340.003242BPBO vs. PKO0.8485-0.67602.37290.5344BPBO vs. SSO0.4521-1.07241.97660.9224NRBO vs. MNRBO7.97806.45359.50243.1894 × 10–25NRBO vs. PKO6.79755.27308.32199.8382 × 10–21NRBO vs. SSO6.40114.87667.92563.4117 × 10–19MNRBO vs. PKO-1.1805-2.70500.34400.2066MNRBO vs. SSO-1.5769-3.1013-0.05240.038916PKO vs. SSO-0.3964-1.92081.12810.9507 Tukey–kramer post-hoc multiple comparison results for Case 2. Moreover, Table 10 shows the Friedman nonparametric statistical test that reveals that the compared optimization algorithms exhibit statistically significant differences in performance, as indicated by the extremely small p-value of 6.041 × 10–13, which is substantially lower than the significance threshold of 0.05. The proposed MNRBO outperforms other algorithms in terms of robustness with a mean rank of 1.00, signifying superior optimization performance. In comparison, the original NRBO is ranked worst with 4.95, highlighting its poor convergence quality and instability, while PKO, SSO, and BPBO hold intermediate ranks. The robustness analysis further highlights the superiority of the proposed MNRBO. It achieves the smallest variance value of 5.022689 × 10–2 and the lowest coefficient of variation (CV = 0.0003), indicating highly stable and consistent performance with minimal fluctuations across repeated simulations. Conversely, NRBO exhibits the highest variance and CV values, reflecting unstable convergence behavior and sensitivity to random initialization. Moreover, the bootstrap confidence intervals provide additional statistical evidence regarding the reliability of the obtained solutions. The proposed MNRBO produces the narrowest confidence interval, ranging from 833.0645 to 833.2512, which indicates highly concentrated and repeatable optimization results. On the other hand, NRBO presents the widest confidence interval, confirming significant variability and reduced robustness. Thus, the combined Friedman ranking, robustness metrics, and bootstrap confidence interval analysis collectively verify that the proposed MNRBO significantly enhances optimization stability, consistency, and convergence reliability compared with the original NRBO and other recent optimization algorithms. Table 10Friedman ranking, robustness analysis, and bootstrap confidence intervals of the compared algorithms for Case 2.AlgorithmMean rankVarianceCV95% bootstrapconfidence intervalof mean objectiveBPBO3.252.837785 × 1000.0020[834.5243, 835.9568]NRBO4.951.107102 × 1010.0040[839.7012, 842.5344]Proposed MNRBO1.005.022689 × 10–20.0003[833.0645, 833.2512]PKO2.804.500142 × 10–10.0008[834.0489, 834.6230]SSO3.006.170659 × 10–10.0009[834.4178, 835.0777]Friedman test resultStatistical testP-valueFriedman test6.041 × 10–13 Table 10Friedman ranking, robustness analysis, and bootstrap confidence intervals of the compared algorithms for Case 2.AlgorithmMean rankVarianceCV95% bootstrapconfidence intervalof mean objectiveBPBO3.252.837785 × 1000.0020[834.5243, 835.9568]NRBO4.951.107102 × 1010.0040[839.7012, 842.5344]Proposed MNRBO1.005.022689 × 10–20.0003[833.0645, 833.2512]PKO2.804.500142 × 10–10.0008[834.0489, 834.6230]SSO3.006.170659 × 10–10.0009[834.4178, 835.0777]Friedman test resultStatistical testP-valueFriedman test6.041 × 10–13 Friedman ranking, robustness analysis, and bootstrap confidence intervals of the compared algorithms for Case 2. Case study 3: fuel cost minimization with multi-fuel characteristics For this case, solving the OPF is addressed with multi-fuel characteristics where Table 11 presents the optimal operating settings obtained by BPBO, NRBO, PKO, SSO, and the proposed MNRBO. Also, Fig. 9 illustrates the best and mean convergence curves for all algorithms. As shown, the MNRBO’s solution yields the lowest fuel cost among all algorithms, achieving 646.7722 $/h, whereas the closest competitor (PKO) produces 647.7604 $/h, and the worst-performing method (NRBO) reaches 649.6727 $/h. This clear improvement highlights MNRBO’s enhanced exploitation ability under discontinuous search landscapes. Fig. 9Convergence of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 3. (a) Best Convergences, (b) Average Convergences. Convergence of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 3. (a) Best Convergences, (b) Average Convergences. Table 11Optimum values of control variables for case 3.S. No.Control variableBPBONRBOProposed MNRBOPKOSSOActive power output of the generator (MW) 1Pg254.999315554.9999354.9818254.96906 2Pg528.3807126.2517622.2494621.8078627.20112 3Pg825.8164730.0755434.783234.9014929.422 4Pg1119.8319.7660916.7583821.500520.3974 5Pg1321.0891919.5597221.339117.2780918.26659Generator voltage 6Vg11.0999931.0726921.0912691.099841.087911 7Vg21.0892011.0693571.0725281.0893061.071947 8Vg51.0633451.0300731.0480521.0977931.035258 9Vg81.0612941.016871.058311.0822371.054088 10Vg111.0540411.0476131.0858891.0893211.034919 11Vg131.028221.0504231.0742311.0949261.022939Tap-setting of transformer 12T6−90.9956321.0056631.0446691.0656681.025501 13T6−100.9916420.9704880.975990.9296521.008853 14T4−120.9781051.0792951.0519631.0998620.995167 15T28−270.9983010.9755080.9971081.0313581.046754Shunt compensator 16Qc102.218592.0575634.0291022.3690110.495302 17Qc122.111462.5988722.9734514.1469692.863469 18Qc152.0747622.1587071.847530.5773110.264317 19Qc172.3078231.4910072.2370690.0217213.015335 20Qc202.6896232.7180733.7832960.6312584.814589 21Qc212.4075142.8939681.5886724.9993031.295095 22Qc231.9756651.5844261.8856080.7286253.553741 23Qc242.3037324.18593.192113.859013.023977 24Qc292.5776740.9542282.0183010.9994151.803163QFC ($/h)649.6054649.6727646.7722647.7604648.7034 Table 11Optimum values of control variables for case 3.S. No.Control variableBPBONRBOProposed MNRBOPKOSSOActive power output of the generator (MW) 1Pg254.999315554.9999354.9818254.96906 2Pg528.3807126.2517622.2494621.8078627.20112 3Pg825.8164730.0755434.783234.9014929.422 4Pg1119.8319.7660916.7583821.500520.3974 5Pg1321.0891919.5597221.339117.2780918.26659Generator voltage 6Vg11.0999931.0726921.0912691.099841.087911 7Vg21.0892011.0693571.0725281.0893061.071947 8Vg51.0633451.0300731.0480521.0977931.035258 9Vg81.0612941.016871.058311.0822371.054088 10Vg111.0540411.0476131.0858891.0893211.034919 11Vg131.028221.0504231.0742311.0949261.022939Tap-setting of transformer 12T6−90.9956321.0056631.0446691.0656681.025501 13T6−100.9916420.9704880.975990.9296521.008853 14T4−120.9781051.0792951.0519631.0998620.995167 15T28−270.9983010.9755080.9971081.0313581.046754Shunt compensator 16Qc102.218592.0575634.0291022.3690110.495302 17Qc122.111462.5988722.9734514.1469692.863469 18Qc152.0747622.1587071.847530.5773110.264317 19Qc172.3078231.4910072.2370690.0217213.015335 20Qc202.6896232.7180733.7832960.6312584.814589 21Qc212.4075142.8939681.5886724.9993031.295095 22Qc231.9756651.5844261.8856080.7286253.553741 23Qc242.3037324.18593.192113.859013.023977 24Qc292.5776740.9542282.0183010.9994151.803163QFC ($/h)649.6054649.6727646.7722647.7604648.7034 Optimum values of control variables for case 3. Figure 10 provides a box-plot comparison of the 20-run statistical performance while Table 12 summarizes key performance indicators for all algorithms across 20 runs. As shown, the MNRBO surpasses all competitors by a significant margin as it clearly demonstrates the most reliable and repeatable performance under multi-fuel discontinuities. Based on the proposed MNRBO methodology, the minimum operational costs have been decreased to 646.7722 $/h. This indicates a reduction of 2.8333 $/h, which corresponds to a 0.44% decrease compared to the BPBO. In comparison with the NRBO, the reduction is 2.9006 $/h corresponds to a 0.45% lower, while the costs are also 0.9882 $/h with a 0.15% less than those with the PKO, and 1.9313 $/h with a 0.3% lower than the SSO. For the average QFC, MNRBO demonstrates the highest statistical performance at 650.0346 $/h, surpassing all alternative methods by a significant margin. The MNRBO achieves reductions of 3.4212 $/h with a 0.52% reduction compared to BPBO, 10.7818 $/h with a 1.63% reduction relative to NRBO, and decreases of 0.9304 $/h with a 0.14% and 1.2479 $/h with a 0.19% versus PKO and SSO respectively. This establishes MNRBO as the most effective method for minimizing costs. For the maximum QFC, the improvements are even more striking. The maximum fuel cost obtained by MNRBO is 653.4908 $/h, whereas NRBO reaches an excessively high 730.7043 $/h, indicating severe instability. Relative to competing methods, MNRBO achieves 1.72%, 10.57%, 0.19%, and 0.12% reductions versus BPBO, NRBO, PKO, and SSO, respectively. This considerable decrease in the maximum cost shows that MNRBO effectively suppresses worst-case divergence, a crucial requirement in multi-fuel OPF where algorithms often get trapped in poor local optima. The findings from this case collectively demonstrate that the proposed MNRBO performs significantly better than NRBO and all competing algorithms for multi-fuel OPF. Fig. 10Boxplots of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 3. Boxplots of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 3. Table 12Statistical indicators of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 3.Case 3BPBONRBOProposedMNRBOPKOSSOMin QFC649.6054198649.6727183646.7721531647.7603517648.7034472Average QFC653.4557698660.8163946650.0345535650.964917651.2824848Max QFC664.9427718730.7043026653.4908319654.7643219654.2433249STD3.17865146716.716615442.316773521.8011753131.739742345 Table 12Statistical indicators of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 3.Case 3BPBONRBOProposedMNRBOPKOSSOMin QFC649.6054198649.6727183646.7721531647.7603517648.7034472Average QFC653.4557698660.8163946650.0345535650.964917651.2824848Max QFC664.9427718730.7043026653.4908319654.7643219654.2433249STD3.17865146716.716615442.316773521.8011753131.739742345 Statistical indicators of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 3. Case study 4: QFC Minimization with multi-fuel characteristics and Prohibited Operating Zones (POZs) For this case, POZs add discontinuities in generator input–output characteristics, where these constraints typically increase the likelihood of optimization stagnation and entrapment in infeasible or suboptimal regions. In this regard, Table 13 lists the optimal dispatch, voltage setpoints, transformer taps, and reactive compensation obtained by the BPBO, NRBO, MNRBO, PKO, and SSO. The proposed MNRBO attains the lowest minimum cost of 674.9264 $/h, outperforming BPBO, NRBO, MNRBO, PKO, and SSO in terms of solution quality which obtain 675.6955, 678.7299, 675.4392, and 677.8257 $/h, respectively. This demonstrates that MNRBO not only consistently avoids POZ-infeasible regions but also locates high-quality feasible optima better than other advanced optimizers. Figure 11 visually compares the statistical spread of QFC solutions for all algorithms under POZ constraints. The boxplot for MNRBO indicates better stable performance across all independent runs. Compared to BPBO and NRBO, the MNRBO shows smaller and most compact, with shorter whiskers across all independent runs. In contrast, NRBO exhibits the widest box and the longest whiskers, demonstrating extreme sensitivity to the feasible regions introduced by POZs. BPBO and PKO show intermediate variability, with scattered outliers reflecting difficulty in escaping infeasible or near-boundary regions. Statistically, Table 14 displays the regarding algorithmic indicators for Case 4. In terms of the average cost comparison, the average cost obtained by MNRBO is 679.2644 $/h, with dramatic improvements over all other methods of 1.88%, 2.47%, 0.08%, and 0.23% lower than BPBO, NRBO, PKO, and SSO. These reductions clearly show that MNRBO maintains exceptional performance across repeated trials, rather than improving only single-run results. Considering the worst-case performance, the maximum cost reported by MNRBO is 682.2787 $/h, significantly better than others, especially NRBO, which shows severe divergence. The proposed MNRBO reduces the maximum costs by 53.5215 $/h, 62.0671 $/h, 0.4289 $/h, and 7.8492 $/h with 7.27%, 8.34%, 0.06%, 1.14% improvements over BPBO, NRBO, PKO, and SSO, respectively. The extremely large gap between MNRBO and NRBO indicates that the newly introduced adaptive mechanisms dramatically enhance NRBO’s ability to avoid unstable oscillatory behavior caused by POZ discontinuities. This confirms that MNRBO provides the most reliable, least fluctuating, and most consistent search behavior under POZ-constrained OPF conditions. Table 13Optimum values of control variables for case 4.S. No.Control variableBPBONRBOProposed MNRBOPKOSSOActive power output of the generator (MW) 1Pg244.9995544.9999944.9971544.9066144.87748 2Pg525.5321521.9191628.2086428.2561827.48763 3Pg830.6841225.7626334.9988231.8410633.01332 4Pg1118.6375923.5527823.4290923.7428622.77835 5Pg1328.4409127.376418.5533921.4703522.78641Generator voltage 6Vg11.0964681.0798421.053591.0916151.04009 7Vg21.0796051.0649471.0381021.0748071.01782 8Vg51.056971.0230761.0153061.0474560.970922 9Vg81.0633761.0400221.0262131.0724070.997305 10Vg111.066580.9959751.0866441.0738321.030668 11Vg131.0564020.9974870.9970540.990841.064801Tap-setting of transformer 12T6−91.043351.0594990.9945931.0849240.967079 13T6−101.0599190.9792041.0896290.9558890.958355 14T4−121.0784860.993391.0723551.0398030.967816 15T28−271.0414311.0087141.0365941.0339331.029746Shunt compensator 16Qc102.6900612.4167180.0013571.1995682.492028 17Qc121.2415313.6009132.4744680.0308710.772161 18Qc153.3704673.6416421.8817940.0456411.897697 19Qc171.0722082.1026714.9027974.7614742.562291 20Qc203.4186771.3810783.2599010.6996862.271801 21Qc213.2464411.0581650.3637341.7544823.961624 22Qc233.3424251.3301451.0886930.7528512.658302 23Qc243.9312660.8515633.3446941.6026812.704271 24Qc291.3962941.7255933.5579234.9688512.387318QFC ($/h)675.6955121678.7299245674.9263608675.4392372677.825784 Table 13Optimum values of control variables for case 4.S. No.Control variableBPBONRBOProposed MNRBOPKOSSOActive power output of the generator (MW) 1Pg244.9995544.9999944.9971544.9066144.87748 2Pg525.5321521.9191628.2086428.2561827.48763 3Pg830.6841225.7626334.9988231.8410633.01332 4Pg1118.6375923.5527823.4290923.7428622.77835 5Pg1328.4409127.376418.5533921.4703522.78641Generator voltage 6Vg11.0964681.0798421.053591.0916151.04009 7Vg21.0796051.0649471.0381021.0748071.01782 8Vg51.056971.0230761.0153061.0474560.970922 9Vg81.0633761.0400221.0262131.0724070.997305 10Vg111.066580.9959751.0866441.0738321.030668 11Vg131.0564020.9974870.9970540.990841.064801Tap-setting of transformer 12T6−91.043351.0594990.9945931.0849240.967079 13T6−101.0599190.9792041.0896290.9558890.958355 14T4−121.0784860.993391.0723551.0398030.967816 15T28−271.0414311.0087141.0365941.0339331.029746Shunt compensator 16Qc102.6900612.4167180.0013571.1995682.492028 17Qc121.2415313.6009132.4744680.0308710.772161 18Qc153.3704673.6416421.8817940.0456411.897697 19Qc171.0722082.1026714.9027974.7614742.562291 20Qc203.4186771.3810783.2599010.6996862.271801 21Qc213.2464411.0581650.3637341.7544823.961624 22Qc233.3424251.3301451.0886930.7528512.658302 23Qc243.9312660.8515633.3446941.6026812.704271 24Qc291.3962941.7255933.5579234.9688512.387318QFC ($/h)675.6955121678.7299245674.9263608675.4392372677.825784 Optimum values of control variables for case 4. Fig. 11Boxplots of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 4. Boxplots of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 4. Table 14Statistical indicators of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 4.Case 4BPBONRBOProposedMNRBOPKOSSOMin QFC675.6955678.7299674.9264675.4392677.8258Average QFC692.2403696.4561679.2644679.798680.8412Max QFC735.8002744.3458682.2787682.7076690.1279STD20.1423820.620212.6791842.0746023.588692 Table 14Statistical indicators of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 4.Case 4BPBONRBOProposedMNRBOPKOSSOMin QFC675.6955678.7299674.9264675.4392677.8258Average QFC692.2403696.4561679.2644679.798680.8412Max QFC735.8002744.3458682.2787682.7076690.1279STD20.1423820.620212.6791842.0746023.588692 Statistical indicators of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 4. Case study 5: RPL minimization For Case 5, the performance of the proposed MNRBO is evaluated under three different system loading conditions that represent real grid operational scenarios: Case 5a (Base Load), Case 5b (Peak Load, 115%), and Case 5c (Off-Peak Load, 85%). In this case, peak and off-peak loading conditions are represented by standardized load-scaling factors applied to the base-case demand, with peak load at 115% and off-peak at 85% of nominal system demand. This method is commonly used in OPF studies to evaluate optimizer performance under varying loads without time-coupled constraints, emphasizing the algorithm’s robustness and adaptability rather than chronological modeling. Table 15 presents the optimal generator dispatch, voltage setpoints, transformer tap ratios, and shunt compensator settings obtained using NRBO and the proposed MNRBO for all three loading levels. Across all operating conditions, the proposed MNRBO demonstrates more coordinated control actions than NRBO, particularly in tap ratios and capacitor settings, highlighting its superior sensitivity to reactive power management and loss-driven optimization. Under base load, MNRBO reduces losses from 2.9228 MW to 2.8517 MW (2.43% reduction). In peak load, it lowers losses from 5.2558 MW to 5.0416 MW (4.07% improvement) by effectively redistributing generation and tightening voltage profiles. The most significant enhancement occurs under off-peak load, where losses drop from 1.8591 MW to 1.7145 MW (7.78% reduction) through precise modulation of shunt compensation and tap changers. These findings confirm MNRBO’s superior ability to minimize losses while ensuring optimal voltage and reactive power coordination across varying conditions. Figure 12 illustrates the convergence performance of MNRBO, NRBO, BPBO, PKO, and SSO under three conditions. MNRBO demonstrates the fastest and smoothest convergence, achieving a lower loss level more quickly than the other algorithms. In contrast, NRBO has a slower convergence and stabilizes at higher loss values, while BPBO, PKO, and SSO show moderate effectiveness but do not reach the convergence speed or precision of MNRBO. Figure 13 highlights the variability of RPL values across 20 runs. In Case 5a, MNRBO shows the tightest box and smallest whisker range, indicating excellent repeatability, while NRBO exhibits greater variation. For Case 5b under peak load, all algorithms show increased variability, but MNRBO remains the most stable. In Case 5c with off-peak load, MNRBO again showcases significant robustness, with much shorter whiskers compared to BPBO, PKO, SSO, and NRBO. The boxplots across these conditions confirm MNRBO’s superior numerical stability despite varying system loads. Table 15Optimum values of control variables for case 5.S. No.Control variableCase 5a (Base Load)Case 5b (Peak Load – 115%)Case 5c (Off-Peak Load – 85%)NRBOProposed MNRBONRBOProposed MNRBONRBOProposed MNRBOActive power output of the generator (MW) 1Pg28079.99999808044.3162737.72872 2Pg55049.9999849.9999949.9999749.8855549.99279 3Pg83534.9998834.99993353534.99553 4Pg113029.999429.996363029.87529.9774 5Pg134039.9978339.999914033.6722439.90999Generator voltage 6Vg11.11.11.11.11.11.1 7Vg21.11.09771.0955041.0942531.0966181.095618 8Vg51.0823731.0799661.0797421.0717061.0824271.083186 9Vg81.090391.0866731.11.0800491.0916211.090291 10Vg111.0668391.11.11.11.0967751.1 11Vg131.11.0999991.11.11.11.099977Tap-setting of transformer 12T6−91.11.0690940.9744211.0503761.054751.078746 13T6−100.90.900021.10.9000011.0143090.900023 14T4−121.0189520.9884381.0905640.9842171.0423810.983739 15T28−271.0071460.9727851.0213110.965380.9894120.979286Shunt compensator 16Qc1054.9935214.9999974.9998454.9875094.999713 17Qc1204.99622553.7864071.306828 18Qc150.75320154.9999854.9995434.4884744.989724 19Qc1754.9997584.999124.9996943.2022683.441065 20Qc2054.999994.999184.9999984.9672222.982745 21Qc2154.9996854.9999934.9999682.7385764.999789 22Qc231.2464172.2415814.9999854.2236434.9844711.259135 23Qc2454.9948374.99995654.2008594.996195 24Qc2952.1794264.9999992.9295564.9999992.193338RPL (MW)2.9228832.8517175.2557735.0416531.8590651.714479 Table 15Optimum values of control variables for case 5.S. No.Control variableCase 5a (Base Load)Case 5b (Peak Load – 115%)Case 5c (Off-Peak Load – 85%)NRBOProposed MNRBONRBOProposed MNRBONRBOProposed MNRBOActive power output of the generator (MW) 1Pg28079.99999808044.3162737.72872 2Pg55049.9999849.9999949.9999749.8855549.99279 3Pg83534.9998834.99993353534.99553 4Pg113029.999429.996363029.87529.9774 5Pg134039.9978339.999914033.6722439.90999Generator voltage 6Vg11.11.11.11.11.11.1 7Vg21.11.09771.0955041.0942531.0966181.095618 8Vg51.0823731.0799661.0797421.0717061.0824271.083186 9Vg81.090391.0866731.11.0800491.0916211.090291 10Vg111.0668391.11.11.11.0967751.1 11Vg131.11.0999991.11.11.11.099977Tap-setting of transformer 12T6−91.11.0690940.9744211.0503761.054751.078746 13T6−100.90.900021.10.9000011.0143090.900023 14T4−121.0189520.9884381.0905640.9842171.0423810.983739 15T28−271.0071460.9727851.0213110.965380.9894120.979286Shunt compensator 16Qc1054.9935214.9999974.9998454.9875094.999713 17Qc1204.99622553.7864071.306828 18Qc150.75320154.9999854.9995434.4884744.989724 19Qc1754.9997584.999124.9996943.2022683.441065 20Qc2054.999994.999184.9999984.9672222.982745 21Qc2154.9996854.9999934.9999682.7385764.999789 22Qc231.2464172.2415814.9999854.2236434.9844711.259135 23Qc2454.9948374.99995654.2008594.996195 24Qc2952.1794264.9999992.9295564.9999992.193338RPL (MW)2.9228832.8517175.2557735.0416531.8590651.714479 Optimum values of control variables for case 5. Fig. 12Average convergences of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 5. (a) Case 5a, (b) Case 5b, (c) Case 5c. Average convergences of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 5. (a) Case 5a, (b) Case 5b, (c) Case 5c. Fig. 13Boxplots of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 5. (a) Case 5a, (b) Case 5b, (c) Case 5c. Boxplots of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 5. (a) Case 5a, (b) Case 5b, (c) Case 5c. In a quantitative way, Table 16 provides a statistical evaluation of algorithms under three loading conditions, demonstrating that the MNRBO algorithm significantly outperforms the NRBO in robustness. It consistently shows the lowest real power losses and superior stability across base, peak, and off-peak loading scenarios. Specifically, under base load (Case 5a), MNRBO reduces minimum loss to 2.8517 MW (2.44% lower than NRBO), cuts average loss by 7.95%, and decreases maximum loss by 11.39%, showing an 82% improvement in standard deviation. In peak load (Case 5b), MNRBO achieves a 4.07% reduction in minimum loss, a 7.92% decrease in average loss, and a 10.68% lower maximum loss, along with a 74.9% enhancement in stability. The most significant gains occur under off-peak conditions (Case 5c), where it reduces minimum loss by 7.78%, average loss by 17.40%, and maximum loss by 29.10%, achieving an impressive 84.56% decrease in variability. These results affirm that MNRBO delivers consistently superior and more reliable performance, excelling especially in reactive-dominant operating environments. For RPL minimization in Case 5a (base load), Table 17 provides a comparative assessment using the proposed MNRBO and several well-established optimization methods. The proposed MNRBO achieves the smallest loss value of 2.8517 MW, outperforming all competing algorithms in terms of solution quality and robustness. Against classical and widely used techniques such as the electromagnetism-like mechanism (2.8699 MW) and the gray wolf optimizer (2.9377 MW), the improvements reach 0.63% and 2.93%, respectively. MNRBO also significantly surpasses evolutionary-based approaches such as DE (3.0850 MW) and improved moth-flame optimization (3.0905 MW), yielding reductions of 7.56% and 7.73%. The results highlight that MNRBO not only outperforms recent metaheuristics (BPBO, PKO, SSO) but also consistently achieves better loss minimization than several well-known optimizers reported in the literature, confirming its superior capability for OPF tasks. Table 16Statistical indicators of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 5.Case studyValueBPBONRBOProposedMNRBOPKOSSOCase 5a(Base Load)Min QFC2.8562042.9228832.8517172.8739142.922219Average QFC2.9405193.1227762.8745622.9042133.017559Max QFC3.1625663.2812182.9080772.9491223.194233STD0.0832830.101770.0183610.0224520.073169Case 5b(Peak Load – 115%)Min QFC5.068925.2557735.0416535.0893695.144728Average QFC5.1939725.5002185.0646155.1775655.27585Max QFC5.3640445.7929815.173395.3180715.415108STD0.0794290.126960.0318110.0651550.078005Case 5c(Off-Peak Load – 85%)Min QFC1.7944861.8590651.7144791.7870911.768465Average QFC1.9558762.1248651.755221.9629851.910403Max QFC2.2317182.5922091.8389792.1440972.169644STD0.1481360.214020.0330660.1115290.095868 Table 16Statistical indicators of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 5.Case studyValueBPBONRBOProposedMNRBOPKOSSOCase 5a(Base Load)Min QFC2.8562042.9228832.8517172.8739142.922219Average QFC2.9405193.1227762.8745622.9042133.017559Max QFC3.1625663.2812182.9080772.9491223.194233STD0.0832830.101770.0183610.0224520.073169Case 5b(Peak Load – 115%)Min QFC5.068925.2557735.0416535.0893695.144728Average QFC5.1939725.5002185.0646155.1775655.27585Max QFC5.3640445.7929815.173395.3180715.415108STD0.0794290.126960.0318110.0651550.078005Case 5c(Off-Peak Load – 85%)Min QFC1.7944861.8590651.7144791.7870911.768465Average QFC1.9558762.1248651.755221.9629851.910403Max QFC2.2317182.5922091.8389792.1440972.169644STD0.1481360.214020.0330660.1115290.095868 Statistical indicators of MNRBO, NRBO, BPBO, PKO and SSO algorithms for Case 5. Case 5a (Base Load) Case 5b (Peak Load – 115%) Case 5c (Off-Peak Load – 85%) Table 17Assessment of OPF results obtained By MNRBO and other reported algorithms For RPL minimization (Case 5a).AlgorithmsRPL (MW)Proposed MNRBO2.851717BPBO2.856204electromagnetism-like mechanismmethod612.8699PKO2.873914SSO2.922219NRBO2.922883Gray wolf optimizer622.9377DE563.0850Improved moth-flame optimization573.0905Moth swarm algorithm603.1005 Table 17Assessment of OPF results obtained By MNRBO and other reported algorithms For RPL minimization (Case 5a).AlgorithmsRPL (MW)Proposed MNRBO2.851717BPBO2.856204electromagnetism-like mechanismmethod612.8699PKO2.873914SSO2.922219NRBO2.922883Gray wolf optimizer622.9377DE563.0850Improved moth-flame optimization573.0905Moth swarm algorithm603.1005 Assessment of OPF results obtained By MNRBO and other reported algorithms For RPL minimization (Case 5a). electromagnetism-like mechanism method61 Case study 6: QFC minimization in the presence of PV systems To more convincingly demonstrate the superiority of the proposed NRBO algorithm, this study is extended by incorporating renewable solar energy sources, along with conventional thermal generators. In this regard, a PV system of 20 MW capacity is considered to be installed at bus 30. In this context, the uncertainties associated with solar photovoltaic (PV) generation are modeled using the Point Estimate Method (PEM), which provides an efficient probabilistic representation of renewable generation variability. A lognormal probability distribution is adopted to characterize the stochastic behavior of solar irradiance as shown in Fig. 14. The probabilistic representation is validated through comparison between the analytical probability density function (PDF) and Monte Carlo Simulation results, where the PEM concentration points and mean value are identified for enhanced visualization and interpretation. Accordingly, the solar irradiance PDF is expressed as:39\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_{Ir} \left( {Ir} \right) = \frac{1}{{Ir \times \sigma_{Ir} \times \sqrt {2\pi } }}e^{{\left( { - \frac{{\left( {\ln \left( {Ir} \right) - \mu_{Ir} } \right)^{2} }}{{2\left( {\sigma_{Ir} } \right)^{2} }}} \right)}}$$\end{document} where; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_{Ir} \left( {Ir} \right)$$\end{document}, solar irradiance PDF. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Ir$$\end{document}, solar irradiance. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu_{Ir }$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma_{Ir}$$\end{document}, mean and standard deviation of the logarithmic irradiance. Based on the PEM framework, two concentration points together with a mean point are utilized to accurately capture the stochastic characteristics of solar PV generation follows63. The concentration points are determined as follows:40\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi_{Ir,k} = \frac{{S_{Ir} }}{2} + \left( { - 1} \right)^{3 - k} \sqrt {\left( {\frac{{K_{Ir} - 3S_{Ir}^{2} }}{4}} \right)} { , }\forall {\text{ k = 1, 2}}$$\end{document} where; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{{Ir,k}}$$\end{document}, concentrate points of solar irradiance. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{{Ir}}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K_{{Ir}}$$\end{document}, skewness and kurtosis of the irradiance distribution. Fig. 14Solar irradiance via Lognormal PDF. Solar irradiance via Lognormal PDF. Consequently, the actual irradiance concentration locations are calculated as:41\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G_{k} = \mu_{Ir} + \left( {\xi_{Ir,k} \times \sigma_{Ir} } \right){ , }\forall {\text{ k = 1, 2}}$$\end{document} To estimate the probabilistic PV power generation, an irradiance-to-power conversion model is employed to transform the stochastic solar irradiance into the corresponding electrical power output of the PV64. The PV generation model is expressed as:42\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_{PV,s} \left( {Ir} \right) = \left\{ {\begin{array}{*{20}c} {P_{PV\_Rated} \times Ir^{2} /\left( {Ir_{c} \times Ir_{STD} } \right)} \\ {P_{PV\_Rated} \times \left( {Ir/Ir_{STD} } \right)} \\ \end{array} } \right. \begin{array}{*{20}c} {0 < Ir < Ir_{c} } \\ {Ir \ge Ir_{c} } \\ \end{array} { }$$\end{document} where; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_{PV,s} \left( {Ir} \right)$$\end{document}, electrical output power corresponding to random irradiance. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Ir_{c}$$\end{document}, Critical irradiance threshold. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Ir_{STD }$$\end{document}, Standard irradiance. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_{PV\_Rated}$$\end{document}, Rated size of the PV system to be installed. To satisfy the moment-matching conditions while preserving the total probability mass, suitable weighting coefficients are assigned to the PEM concentration points. Multiple uncertainty scenarios are subsequently generated to construct a combined point-estimate matrix representing the stochastic PV generation behavior. For this case study, the optimization objective considers probabilistic fuel cost minimization over all PEM scenarios. Therefore, the total objective function is formulated as the weighted summation of the quadratic fuel costs corresponding to all identified PEM operating scenarios: 43\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f = \mathop \sum \limits_{s = 1}^{2m + 1} \left( {\omega_{s} \times QFC_{s} } \right),{\text{ m = 2 variables}}$$\end{document} The integration of solar PV energy is investigated in this case study under uncertain irradiance conditions. The solar irradiance uncertainty is modeled using a lognormal probability distribution characterized by a mean value µIr of 6 and a σIr of 0.665. By considering the number of concentration points used is three and applying the PEM, three representative irradiance scenarios of 252.38, 482.99 and 1194.36 W/m2 are generated, corresponding to low, medium, and high solar irradiance conditions, respectively. Table 18 summarizes the optimal OPF control variable settings obtained using the original NRBO and the proposed MNRBO algorithms. The obtained results demonstrate that the proposed MNRBO achieves a lower expected quadratic fuel cost (QFC) of 1004.335 $/h compared with 1007.249 $/h obtained by NRBO, representing an improvement of approximately 0.29%. Although the numerical reduction appears moderate, it reflects enhanced optimization efficiency considering the highly constrained probabilistic OPF environment and renewable uncertainty modeling. The best and average convergence characteristics of NRBO and MNRBO are illustrated in Fig. 15. The proposed MNRBO exhibits faster convergence behavior and improved stability compared with NRBO. Table 18Optimum values of control variables for case 6.S. no.Control variableNRBOProposed MNRBOActive power output of the generator (MW) 1Pg244.4738947.53447 2Pg520.1470420.94524 3Pg820.6428818.23136 4Pg1114.6380910.97053 5Pg1312.0805612Generator voltage 6Vg11.11.1 7Vg21.0870951.087892 8Vg51.0562561.061785 9Vg81.0572081.069976 10Vg111.09281.1 11Vg131.0957291.1Tap-setting of transformer 12T6−91.0294041.066088 13T6−101.0599790.900139 14T4−121.0089020.995698 15T28−271.0801690.973096Shunt compensator 16Qc101.0833344.999931 17Qc122.6170844.999046 18Qc151.7011814.144607 19Qc171.9392624.992496 20Qc201.1416224.791963 21Qc213.2605334.995833 22Qc233.407762.770972 23Qc240.5764224.98959 24Qc293.5703880.736789Expected QFC ($/h)1007.2492411004.334971 Table 18Optimum values of control variables for case 6.S. no.Control variableNRBOProposed MNRBOActive power output of the generator (MW) 1Pg244.4738947.53447 2Pg520.1470420.94524 3Pg820.6428818.23136 4Pg1114.6380910.97053 5Pg1312.0805612Generator voltage 6Vg11.11.1 7Vg21.0870951.087892 8Vg51.0562561.061785 9Vg81.0572081.069976 10Vg111.09281.1 11Vg131.0957291.1Tap-setting of transformer 12T6−91.0294041.066088 13T6−101.0599790.900139 14T4−121.0089020.995698 15T28−271.0801690.973096Shunt compensator 16Qc101.0833344.999931 17Qc122.6170844.999046 18Qc151.7011814.144607 19Qc171.9392624.992496 20Qc201.1416224.791963 21Qc213.2605334.995833 22Qc233.407762.770972 23Qc240.5764224.98959 24Qc293.5703880.736789Expected QFC ($/h)1007.2492411004.334971 Optimum values of control variables for case 6. Fig. 15Convergence of NRBO and MNRBO algorithms for Case 6. (a) Best convergences, (b) average convergences. Convergence of NRBO and MNRBO algorithms for Case 6. (a) Best convergences, (b) average convergences. Initially, load-flow analysis is performed without PV integration in order to establish the base operating condition for comparison purposes. Figure 16 presents the bus voltage profiles under the three PEM scenarios before and after applying the proposed MNRBO. In the initial operating condition, several buses experience noticeable voltage degradation, particularly under low irradiance conditions where the reduced PV contribution increases network stress. After applying MNRBO, the voltage profiles are significantly enhanced across all buses and all PEM scenarios. Most bus voltages are successfully shifted toward the acceptable operational range near 1.0 p.u., indicating improved voltage stability and reactive power management. The proposed MNRBO also demonstrates substantial economic improvements under all irradiance scenarios, as illustrated in Fig. 17. For Scenario 1, the generation cost is reduced from 835.016 $/h to 715.619 $/h, corresponding to an improvement of approximately 14.30%. Under Scenario 2, the cost decreases from 886.046 $/h to 780.460 $/h, yielding an improvement of nearly 11.92%. Similarly, for Scenario 3, the cost is reduced from 872.730 $/h to 763.776 $/h, corresponding to an enhancement of approximately 12.49%. These results confirm that the proposed MNRBO effectively handles the stochastic behavior of renewable PV generation while simultaneously improving system economy, voltage regulation, and convergence robustness under probabilistic operating conditions. Fig. 16Buses voltage for the three PEM scenarios regarding the initial case and MNRBO for Case 6. Buses voltage for the three PEM scenarios regarding the initial case and MNRBO for Case 6. Fig. 17QFC for the three PEM scenarios regarding the initial case and MNRBO for Case 6. (a) Initial Case, (b) MNRBO application. QFC for the three PEM scenarios regarding the initial case and MNRBO for Case 6. (a) Initial Case, (b) MNRBO application. Ablation study for QFC minimization in the presence of PV systems To further investigate the effectiveness of the proposed modifications introduced in the MNRBO framework, an ablation study is conducted for Case 6 under probabilistic PV operating conditions. The objective of this analysis is to quantify the individual and combined contributions of the ACM and the Sigmoid Decay strategy relative to the baseline NRBO algorithm. Four optimization variants are considered in this study: (i) The original NRBO algorithm; (ii) NRBO integrated only with the ACM; (iii) NRBO integrated only with the Sigmoid Decay strategy; and (iv) The complete proposed MNRBO incorporating both ACM and Sigmoid Decay. Table 19 summarizes the statistical performance indicators obtained from multiple independent runs, including the minimum, average, and maximum QFC, in addition to the standard deviation (STD). The results indicate that the ACM significantly enhances optimization quality and convergence robustness. When integrated with the NRBO, the minimum QFC improves from 1007.249 $/h to 1004.387 $/h (approximately 0.28% reduction), and the average QFC decreases from 1013.771 $/h to 1004.472 $/h (about 0.92% enhancement). Additionally, solution stability improves, with the standard deviation falling from 6.47067 to 0.080659, indicating a 98.75% reduction in performance fluctuations, confirming the effectiveness of ACM in enhancing population diversity and preventing premature convergence. Conversely, the variant employing only the sigmoid decay mechanism alone does not significantly enhance optimization, as indicated by a slight increase in average QFC values from 1013.771 $/h to 1014.526 $/h and an increase in standard deviation to 6.848032. This reflects reduced robustness when used without adaptive crossover support, suggesting that its effectiveness relies on sufficient population diversity provided by ACM. The complete MNRBO framework, integrating ACM and sigmoid decay, performs best among variants with a minimum QFC of 1004.335 $/h and an average QFC of 1004.389 $/h, marking a 0.93% improvement over the original NRBO. It also exhibits high robustness, with an STD of 0.023766, indicating a 99.63% reduction compared to NRBO, and shows narrow variation in QFC values, highlighting its consistency and convergence reliability. These findings indicate that the enhanced performance of MNRBO results from the interaction between ACM and Sigmoid Decay. While ACM ensures exploration capability and population diversity, Sigmoid Decay facilitates a smooth transition to exploitation and stable convergence in optimization stages. This integration makes NRBO more adaptive and robust for probabilistic OPF problems involving renewable energy uncertainty. Table 19Ablation analysis of NRBO, NRBO with ACM, NRBO with sigmoid decay and MNRBO for case 6.Case 6NRBONRBO withACM onlyNRBO with sigmoiddecay onlyMNRBOMin QFC1007.2491004.3871008.7131004.335Average QFC1013.7711004.4721014.5261004.389Max QFC1033.7121004.6881032.5391004.423STD6.470670.0806596.8480320.023766 Table 19Ablation analysis of NRBO, NRBO with ACM, NRBO with sigmoid decay and MNRBO for case 6.Case 6NRBONRBO withACM onlyNRBO with sigmoiddecay onlyMNRBOMin QFC1007.2491004.3871008.7131004.335Average QFC1013.7711004.4721014.5261004.389Max QFC1033.7121004.6881032.5391004.423STD6.470670.0806596.8480320.023766 Ablation analysis of NRBO, NRBO with ACM, NRBO with sigmoid decay and MNRBO for case 6. Case study 7: QFC minimization for large scale IEEE 118-bus system The IEEE 118-bus system is considered as a large-scale and highly constrained OPF benchmark consisting of 118 buses, 186 transmission lines, 54 generators, 9 transformer tap changers, and 14 shunt reactive compensators66. In this case study, the QFC minimization problem is solved using both the standard NRBO and the proposed MNRBO algorithms in order to investigate their effectiveness on high-dimensional optimization problems. Figure 18 illustrates the convergence characteristics of both algorithms. The detailed optimal settings of generator voltages, transformer tap positions, reactive compensators, and active power outputs are presented in Tables 20 and 21, and 22. As observed, the proposed MNRBO achieves a lower final objective value compared with the original NRBO. The best QFC obtained by MNRBO reaches 129794.2 $/h, whereas the standard NRBO attains 129,984 $/h, confirming the superior exploitation capability of the proposed modifications in large-scale search spaces. Fig. 18Convergence curves of NRBO and proposed MNRBO for Case 7. Convergence curves of NRBO and proposed MNRBO for Case 7. The robustness and stability of both methods are further investigated through 30 independent runs, as illustrated in Fig. 19 and the boxplot representation in Fig. 20. The statistical results clearly demonstrate the enhanced consistency of the proposed MNRBO. Specifically, MNRBO achieves a minimum QFC of 129794.23 $/h and an average QFC of 129862.5 $/h, both of which are substantially lower than those obtained by NRBO, which records 129984.018 $/h and 130,277 $/h, respectively. Moreover, the maximum obtained QFC is reduced from 130398.9 $/h for NRBO to 130030.2 $/h for MNRBO. In addition, the standard deviation is significantly reduced from 83.99427 in NRBO to 49.59358 in the proposed MNRBO, representing approximately 40.9% improvement in robustness and solution stability. Fig. 19Thirty runs of NRBO and MNRBO for Case 7. Thirty runs of NRBO and MNRBO for Case 7. Fig. 20Boxplot of NRBO and MNRBO for Case 7. Boxplot of NRBO and MNRBO for Case 7. Table 20Optimal voltage settings of the generators of NRBO and proposed MNRBO for Case 7.NRBOMNRBONRBOMNRBONRBOMNRBONRBOMNRBOVg11.05911.0043Vg321.05571.0307Vg661.04141.0584Vg921.05021.0425Vg41.05931.0251Vg341.05941.0137Vg691.05461.0600Vg991.05451.0344Vg61.05681.0214Vg361.05481.0113Vg701.05711.0301Vg1001.05921.0389Vg81.05931.0597Vg401.05720.9960Vg721.05391.0349Vg1031.04861.0328Vg101.05801.0600Vg421.05950.9982Vg731.05521.0280Vg1041.05251.0189Vg121.05441.0218Vg461.03111.0290Vg741.05041.0043Vg1051.04751.0148Vg151.05661.0126Vg491.05651.0413Vg761.04181.0001Vg1071.04321.0065Vg181.05941.0211Vg541.05841.0137Vg771.04051.0371Vg1101.04591.0099Vg191.05541.0119Vg551.05521.0124Vg801.05491.0541Vg1111.05941.0254Vg241.05931.0467Vg561.05531.0123Vg851.05371.0400Vg1121.03091.0024Vg251.05461.0600Vg591.05691.0306Vg871.05671.0262Vg1131.05671.0305Vg261.03991.0600Vg611.05951.0334Vg891.05941.0597Vg1161.05241.0345Vg271.04981.0325Vg621.05361.0301Vg901.05691.0326Vg311.05581.0235Vg651.05791.0481Vg911.05941.0352 Table 20Optimal voltage settings of the generators of NRBO and proposed MNRBO for Case 7.NRBOMNRBONRBOMNRBONRBOMNRBONRBOMNRBOVg11.05911.0043Vg321.05571.0307Vg661.04141.0584Vg921.05021.0425Vg41.05931.0251Vg341.05941.0137Vg691.05461.0600Vg991.05451.0344Vg61.05681.0214Vg361.05481.0113Vg701.05711.0301Vg1001.05921.0389Vg81.05931.0597Vg401.05720.9960Vg721.05391.0349Vg1031.04861.0328Vg101.05801.0600Vg421.05950.9982Vg731.05521.0280Vg1041.05251.0189Vg121.05441.0218Vg461.03111.0290Vg741.05041.0043Vg1051.04751.0148Vg151.05661.0126Vg491.05651.0413Vg761.04181.0001Vg1071.04321.0065Vg181.05941.0211Vg541.05841.0137Vg771.04051.0371Vg1101.04591.0099Vg191.05541.0119Vg551.05521.0124Vg801.05491.0541Vg1111.05941.0254Vg241.05931.0467Vg561.05531.0123Vg851.05371.0400Vg1121.03091.0024Vg251.05461.0600Vg591.05691.0306Vg871.05671.0262Vg1131.05671.0305Vg261.03991.0600Vg611.05951.0334Vg891.05941.0597Vg1161.05241.0345Vg271.04981.0325Vg621.05361.0301Vg901.05691.0326Vg311.05581.0235Vg651.05791.0481Vg911.05941.0352 Optimal voltage settings of the generators of NRBO and proposed MNRBO for Case 7. Table 21Optimal power outputs of the generators of NRBO and proposed MNRBO for Case 7.NRBOMNRBONRBOMNRBONRBOMNRBOPg126.030528.3376Pg4242.112940.3132Pg80420.5836420.4654Pg43.28434.6925Pg4620.147120.2412Pg854.99055.1313Pg65.33722.0081Pg49198.6337191.9020Pg877.00543.2583Pg84.82290.0199Pg5452.560648.4175Pg89481.8693495.1872Pg10384.4259394.4142Pg5529.835937.9512Pg903.77340.4968Pg1288.282485.7844Pg5631.107734.2268Pg915.84144.4333Pg1520.527511.1875Pg59147.6095146.4350Pg923.99063.3686Pg1815.019212.3242Pg61152.3552145.1250Pg995.45930.0056Pg1920.874223.2137Pg625.40007.1518Pg100223.9223227.1915Pg244.59505.3803Pg65341.0513351.8854Pg10338.410338.8640Pg25190.3575192.0782Pg66342.2116344.0024Pg1044.26603.0815Pg26271.8554275.1732Pg69454.5179454.7749Pg10510.08167.8666Pg278.493417.2804Pg705.37341.9996Pg10728.894328.8528Pg319.80858.6271Pg724.92825.0925Pg1108.92298.3651Pg3215.024314.4212Pg735.93483.9741Pg11139.073632.8973Pg346.65605.7307Pg7413.520614.1207Pg11233.370039.1307Pg369.011210.0241Pg7623.845214.0156Pg1134.34990.0121Pg4051.011249.5377Pg776.95220.0514Pg1165.27546.7018 Table 21Optimal power outputs of the generators of NRBO and proposed MNRBO for Case 7.NRBOMNRBONRBOMNRBONRBOMNRBOPg126.030528.3376Pg4242.112940.3132Pg80420.5836420.4654Pg43.28434.6925Pg4620.147120.2412Pg854.99055.1313Pg65.33722.0081Pg49198.6337191.9020Pg877.00543.2583Pg84.82290.0199Pg5452.560648.4175Pg89481.8693495.1872Pg10384.4259394.4142Pg5529.835937.9512Pg903.77340.4968Pg1288.282485.7844Pg5631.107734.2268Pg915.84144.4333Pg1520.527511.1875Pg59147.6095146.4350Pg923.99063.3686Pg1815.019212.3242Pg61152.3552145.1250Pg995.45930.0056Pg1920.874223.2137Pg625.40007.1518Pg100223.9223227.1915Pg244.59505.3803Pg65341.0513351.8854Pg10338.410338.8640Pg25190.3575192.0782Pg66342.2116344.0024Pg1044.26603.0815Pg26271.8554275.1732Pg69454.5179454.7749Pg10510.08167.8666Pg278.493417.2804Pg705.37341.9996Pg10728.894328.8528Pg319.80858.6271Pg724.92825.0925Pg1108.92298.3651Pg3215.024314.4212Pg735.93483.9741Pg11139.073632.8973Pg346.65605.7307Pg7413.520614.1207Pg11233.370039.1307Pg369.011210.0241Pg7623.845214.0156Pg1134.34990.0121Pg4051.011249.5377Pg776.95220.0514Pg1165.27546.7018 Optimal power outputs of the generators of NRBO and proposed MNRBO for Case 7. Table 22Optimal tap settings and reactive power compensations of NRBO and proposed MNRBO for Case 7.Tap settingsReactive power compensationsNRBOMNRBONRBOMNRBONRBOMNRBOT8−50.97921.0161Qc342.75701.5815Qc792.06401.3802T26−251.00951.0009Qc441.95173.5854Qc822.40483.1529T30−171.00501.0248Qc452.51222.7590Qc832.00561.7950T38−370.97941.0160Qc461.17502.5294Qc1052.62062.9719T63−590.96680.9750Qc481.65822.0257Qc1071.45712.1646T64−610.96281.0020Qc743.08041.2510Qc1101.92241.8864T65−661.02491.0021QFC129984.018129794.23T68−690.96880.9351T81−800.99970.9822 Table 22Optimal tap settings and reactive power compensations of NRBO and proposed MNRBO for Case 7.Tap settingsReactive power compensationsNRBOMNRBONRBOMNRBONRBOMNRBOT8−50.97921.0161Qc342.75701.5815Qc792.06401.3802T26−251.00951.0009Qc441.95173.5854Qc822.40483.1529T30−171.00501.0248Qc452.51222.7590Qc832.00561.7950T38−370.97941.0160Qc461.17502.5294Qc1052.62062.9719T63−590.96680.9750Qc481.65822.0257Qc1071.45712.1646T64−610.96281.0020Qc743.08041.2510Qc1101.92241.8864T65−661.02491.0021QFC129984.018129794.23T68−690.96880.9351T81−800.99970.9822 Optimal tap settings and reactive power compensations of NRBO and proposed MNRBO for Case 7. Furthermore, Table 23 compares the proposed MNRBO with several well-established optimization approaches reported in the literature, including Social Spider Optimization (SSO), Improved SSO, Moth-Swarm Algorithm (MSA), enhanced MSA, colliding Bodies Optimization, DE, PSO, and Backtracking Search Optimization (BSO). The proposed MNRBO achieves one of the lowest fuel cost values among all compared methods. In particular, MNRBO improves the QFC compared with NRBO, DE, PSO, MSA, Enhanced MSA, CBO, and BSO by approximately 0.15%, 0.55%, 0.38%, 4.08%, 4.04%, 4.07%, and 4.09%, respectively. Table 23Comparison results for QFC minimization for 118-bus system.MethodCostProposed MNRBO129794.23NRBO129984.018social spider optimization67132080.4118Improved social spider optimization67129879.4536moth-swarm algorithm68135310.84Enhanced moth-swarm algorithm68135262.57Colliding bodies optimization21135297.22DE69130518.5PSO70130288.210backtracking search optimization71135333.5 Table 23Comparison results for QFC minimization for 118-bus system.MethodCostProposed MNRBO129794.23NRBO129984.018social spider optimization67132080.4118Improved social spider optimization67129879.4536moth-swarm algorithm68135310.84Enhanced moth-swarm algorithm68135262.57Colliding bodies optimization21135297.22DE69130518.5PSO70130288.210backtracking search optimization71135333.5 Comparison results for QFC minimization for 118-bus system. Ablation study for QFC minimization in large scale deterministic case of 118-bus system To further investigate the individual and combined contributions of the proposed ACM and sigmoid decay strategy under large-scale optimization conditions, an additional ablation study is conducted on the IEEE 118-bus OPF system. The four algorithmic variants are analyzed and the statistical results from multiple runs are presented in Table 24. The results indicate that the modifications enhance the optimization performance of the NRBO, with the ACM improving solution stability by reducing the standard deviation from 83.994 to 71.529 (approximately 14.84%). Additionally, ACM slightly lowers the average fuel cost from 130165.7 $/h to 130155.7 $/h, reflecting maintained population diversity. However, it does not improve the minimum fuel cost and yields a marginally higher best solution than the original NRBO, showing that enhancing diversity alone does not fully utilize promising areas in a high-dimensional search space. In contrast, the sigmoid decay strategy enhances optimization quality by reducing the minimum QFC from 129984.018 $/h to 129799.4 $/h. The average QFC decreases by 289.4 $/h, and standard deviation is lowered to 60.810, indicating a robustness improvement of approximately 27.6% compared to the original NRBO. Moreover, the proposed MNRBO framework demonstrates the highest performance among the variants studied, achieving a minimum fuel cost of 129794.23 $/h, the lowest objective value in the research. It also reduces the average fuel cost to 129857.1 $/h, an improvement of about 308.6 $/h (0.24%) over the original NRBO. The maximum cost decreases from 130,277.9 $/h to 130,024.6 $/h, indicating enhanced optimization consistency. MNRBO has the lowest standard deviation of 49.594, improving by approximately 40.96% compared to original NRBO, and surpassing ACM-only and sigmoid-decay-only variants by 30.7% and 18.4%, respectively, confirming that these methods alone do not match the full performance of MNRBO. Table 24Ablation analysis of NRBO, NRBO with ACM, NRBO with sigmoid decay and MNRBO for Case 7.Case 7NRBONRBO withACM onlyNRBO with sigmoiddecay onlyMNRBOMin QFC129984.018130022.5129799.4129794.23Average QFC130165.7130155.7129876.3129857.1Max QFC130277.9130241.2130097.7130024.6STD83.9942771.5291260.8097549.59358 Table 24Ablation analysis of NRBO, NRBO with ACM, NRBO with sigmoid decay and MNRBO for Case 7.Case 7NRBONRBO withACM onlyNRBO with sigmoiddecay onlyMNRBOMin QFC129984.018130022.5129799.4129794.23Average QFC130165.7130155.7129876.3129857.1Max QFC130277.9130241.2130097.7130024.6STD83.9942771.5291260.8097549.59358 Ablation analysis of NRBO, NRBO with ACM, NRBO with sigmoid decay and MNRBO for Case 7. For completeness and reproducibility, the computational time was also evaluated for the large-scale IEEE 118-bus system. The average execution times were 509.520 s for the original NRBO and 513.657 s for the proposed MNRBO. The additional computational burden introduced by the proposed modifications is therefore minimal (approximately 0.81%), indicating that the performance improvements obtained by MNRBO are achieved without a significant increase in computational cost. This demonstrates the practical applicability of the proposed method for large-scale OPF problems. Conclusions This study presented an enhanced MNRBO to improve exploration-exploitation balance, increase numerical stability, and accelerate convergence when solving challenging OPF problems. The upgraded MNRBO features two adaptive mechanisms, ACM and SDM, which work together to prevent premature convergence, manage non-smooth cost functions, and dynamically adjust search behavior throughout the optimization process. Seven OPF scenarios were evaluated on the IEEE 30-bus system, covering a broad spectrum of operating conditions under base, peak, and off-peak loading. The proposed MNRBO outperformed the original NRBO and three modern algorithms of BPBO, PKO, and SSO in terms of solution quality and robustness, across all benchmarks, showing superior minimum, mean, and maximum objective values. In cost-minimization, MNRBO reduced average costs by up to 2.2% with substantial stability improvements, surpassing 80–90% reductions in standard deviation compared to NRBO. For loss minimization, it achieved reductions of 2.43% at base load, 4.07% at peak load, and 7.78% during off-peak conditions. Moreover, the proposed MNRBO effectively manages renewable generation uncertainty while enhancing voltage regulation, economic performance, and convergence stability. It combines PEM-based stochastic irradiance modeling with OPF optimization, proving capable in practical renewable-integrated power system optimization. An ablation study indicates that MNRBO’s superior performance stems from the interaction of ACM and sigmoid decay strategies, where ACM boosts exploration and diversity, and sigmoid decay ensures smoother exploitation and convergence. Otherwise, the MNRBO shows high robustness and scalability in the IEEE 118-bus system, achieving lower fuel costs and reduced variability compared to NRBO and other optimization methods, with statistical analyses revealing MNRBO’s consistent reliability and tight performance distribution. Practical considerations From a practical implementation perspective, the steady, dependable convergence of MNRBO highlights the effectiveness of its adaptive mechanisms in navigating complex, multimodal search landscapes. The MNRBO demonstrates strong performance in terms of solution quality and robustness; however, several aspects should be considered. First, the inclusion of the adaptive crossover mechanism and sigmoid-based control strategy introduces a moderate increase in computational effort compared to the original NRBO, representing a trade-off between accuracy and execution time. Second, while the current results on the IEEE 30-bus system validate its effectiveness, extending the method to larger-scale systems may pose additional challenges in terms of scalability and convergence speed. These aspects highlight important directions for future enhancements toward real-world deployment. Future works Although the proposed MNRBO has demonstrated strong performance and robustness across various OPF scenarios, several promising directions remain for future research. First, incorporating renewable energy sources along with uncertainty modeling would enable the evaluation of MNRBO in stochastic and real-time OPF frameworks. Additionally, future work may explore the integration of hydro-thermal coordination and multi-period dynamic OPF formulations to account for time-coupled constraints and storage dynamics. From an algorithmic perspective, adaptive parameter self-tuning and hybridization with machine learning techniques could further enhance convergence speed and solution accuracy. Finally, real-time implementation and hardware-in-the-loop validation may provide deeper insights into the practical deployment of MNRBO in modern smart grid environments.