Leveraging Predictions in Power System Voltage Control: An Adaptive Approach
Wenqi Cui, Yiheng Xie, Steven Low, Adam Wierman, Baosen Zhang · 2025
Read Leveraging Predictions in Power System Voltage Control: An Adaptive Approach by Wenqi Cui, Yiheng Xie, Steven Low, Adam Wierman, Baosen Zhang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-09937.txt.
Introduction
Wenqi Cui and 4 co-authors address voltage control on power distribution networks in this 2025 paper. Solar panels, electric vehicles, and storage make net load swing quickly, so voltage can drift. Most existing controllers assume load stays steady long enough for their actions to settle, an assumption that renewable-heavy grids break more and more often.
Their adaptive design feeds short-term load forecasts, built from local measurements, straight into the controller. They prove the overall setup stays stable in a decentralized way, even as conditions change, and they tune the control policy with reinforcement learning. Tests used time-varying load records from a real distribution system, showing the forecast-aware design in action under realistic swings.
Catalog record
| CREATOR | Wenqi Cui, Yiheng Xie, Steven Low, Adam Wierman, Baosen Zhang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 50 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-09937
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Leveraging Predictions in Power System Voltage Control: An Adaptive Approach Wenqi Cui, Yiheng Xie, Steven Low, Adam Wierman, Baosen Zhang Abstract High variability of solar PV and sudden changes in load (e.g., electric vehicles and storage) can lead to large voltage fluctuations in the distribution system. In recent years, a number of controllers have been designed to optimize voltage control. These controllers, however, almost always assume that the net load in the system remains constant over a sufficiently long time, such that the control actions converge before the load changes again. Given the intermittent and uncertain nature of renewable resources, it is becoming important to explicitly consider net load that is time-varying. This paper proposes an adaptive approach to voltage control in power systems with significant time-varying net load. We leverage advances in short-term load forecasting, where the net load in the system can be partially predicted using local measurements. We integrate these predictions into the design of adaptive controllers, and prove that the overall control architecture achieves input-to-state stability in a decentralized manner. We optimize the control policy through reinforcement learning. Case studies are conducted using time-varying load data from a real-world distribution system. I Introduction The increasing adoption of distributed energy resources (DERs) brings high variability and intermittency in power injections across distribution systems. These fluctuations can lead to rapid and substantial voltage deviations that occur on timescales much shorter than those handled by conventional mechanical control devices such as tap-changing transformers [ 1 , 2 , 3 ] . To address this issue, the fast-acting capabilities of inverter-based resources (IBRs) offer a promising alternative by controlling their reactive power injections in response to voltage deviations [ 4 , 5 , 6 ] . A plethora of control laws have been proposed for voltage regulation, and the most representative designs include linear droop control [ 7 , 8 ] , incremental voltage control law [ 9 , 6 , 10 ] , and learning-based monotone controllers [ 1 , 11 , 12 , 13 ] . A common assumption underlying all of the aforementioned works is the existence of a separation in timescale, where the system has sufficient time to reach a steady state after a disturbance such as a sudden change of load. However, due to the intermittent and uncertain nature of renewable energy sources and larger compute loads, this assumption may no longer hold: net load variations can persist such that the system never reaches the steady state. Therefore, it becomes essential to explicitly model the time-varying characteristics of net load and to design voltage control laws that can effectively respond to these time-varying patterns. One way to handle time variation is to use robust controllers, since they can potentially achieve performance guarantees under uncertainties. To achieve robustness towards the uncertainties in PV generation, [ 14 ] proposes to modulate the smart inverter reactive power as a function of its real power. To address uncertainties over time horizons, a robust constrained model predictive control is constructed for centralized voltage control with respect to uncertain power outputs from DERs [ 15 ] . To mitigate the impact of uncertain grid topologies, [ 16 ] and [ 17 ] propose robust voltage control mechanisms with an unknown grid topology. However, these robust control strategies often rely on real-time wide-area communication infrastructures, which may not be feasible in distribution networks. To address communication limitations, [ 18 ] proposes a distributed voltage control scheme using the alternating direction method of multipliers (ADMM) algorithm. Building upon the design in [ 14 ] , [ 19 ] proposes a data-driven and distributed scheme for robust optimization of voltage control gains, where system information and wide-area communications is not required. However, these approaches still require some form of information exchange among neighboring nodes, which can be difficult to implement in practice. In addition, the robust formulation tends to produce conservative solutions that might be far from effective in typical operating scenarios. Instead of modeling the time-varying load as uncertainties in a bounded region or following certain distributions, recent advances in short-term predictions show that future load or renewables can be largely predicted from historical data [ 20 , 21 ] . This motivates us to leverage local predictions of the net load in voltage control, so that the controller can adapt to time-varying changes of the net load. However, prediction errors are inevitable, and local forecasts do not fully capture the influence of other buses in the network. This gives rise to two central challenges we aim to address: how to systematically embed predictions in control, and what performance guarantees can be established. Fig. 1: Structure of the adaptive approach for voltage control with time-varying net load. We consider predictions as a set of basis functions, and the predictable time-varying load is expressed through a linear combination of the basis functions. We integrate these predictions into the design of adaptive controllers, which provides a flexible way for incorporating diverse prediction models with minimal modifications to existing voltage control laws. The overall control architecture achieves input-to-state stability in a decentralized manner. To address these challenges, this paper proposes to embed predictions in the voltage controller via an adaptive control framework. We consider predictions as a set of basis functions, and the predictable time-varying load is expressed through a linear combination of the basis functions. By designing an adaptive law on the combination coefficients, the predictable part is able to follow the trend of the real-time time-varying net load. This provides a flexible way for incorporating diverse prediction models with minimal modifications to existing voltage control laws. In addition, we show that the closed-loop control guarantees voltage stability despite the existence of prediction errors. Case studies are conducted using both illustrative net load and real-world load data from an operational distribution grid [ 22 ] , demonstrating the effectiveness of the proposed approach in reducing voltage fluctuations and violations. The structure of the proposed approach is illustrated in Fig. 1 . The contributions of this paper is as follows: 1. We propose an adaptive control framework to embed predictions into the voltage controller, which reduces voltage fluctuations in the presence of time-varying load and power injections. 2. We prove that the proposed controller design guarantees the input-to-state stability of the voltage dynamics, where the impact of prediction error on voltage deviations is analytically quantified. 3. The proposed method provides a framework for leveraging abundant offline data for real-time voltage control. Through the adaptive design, the proposed controller could bridge the gap between offline training of predictions and online operations. The remainder of this paper is organized as follows. Section II introduces the problem formulation for voltage control in distribution systems. Section III characterizes the modeling of time-varying net power injections and derives equivalent voltage dynamics for the design of decentralized controllers. Section IV proposes an adaptive approach to embed predictions with minimal modifications to existing voltage control laws. The conditions for achieving exponential input-to-state stability are also derived. Section V shows the simulation results, and Section VI concludes the paper. II Model and Problem Formulation II-A Notation Throughout this manuscript, vectors are denoted in lower-case bold and matrices are denoted in upper-case bold, and scalars