Bayesian and Multi-Objective Decision Support for Real-Time Cyber-Physical Incident Mitigation Journal: Computers & Security
Shaofei Huang, Christopher M. Poskitt, Lwin Khin Shar · 2025
Read Bayesian and Multi-Objective Decision Support for Real-Time Cyber-Physical Incident Mitigation Journal: Computers & Security by Shaofei Huang, Christopher M. Poskitt, Lwin Khin Shar (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-2509-00770.txt.
Introduction
Shaofei Huang, with two co-authors, builds decision support for cyber-physical incidents. Such systems link physical processes with digital control. A security incident there can quickly become a safety failure. Existing tools rely on static views of risk. They also use single-goal risk models that miss key trade-offs.
The new framework joins three parts into one pipeline. One part builds layered probability networks from the system design. A second part estimates exposure with calibrated confidence. A third part searches over sets of fixes for the best trade-offs. Simple frequency rules then rank the actions to take first. Tests use three attack scenarios that stand for real cases. The framework adapts as threats evolve over time.
Catalog record
| CREATOR | Shaofei Huang, Christopher M. Poskitt, Lwin Khin Shar |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 69 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-00770
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Bayesian and Multi-Objective Decision Support for Real-Time Cyber-Physical Incident Mitigation Journal: Computers & Security Shaofei Huang, Christopher M. Poskitt, Lwin Khin Shar Abstract This research proposes a real-time, adaptive decision-support framework for mitigating cyber incidents in cyber-physical systems, developed in response to an increasing reliance on these systems within critical infrastructure and evolving adversarial tactics. Existing decision-support systems often fall short in accounting for multi-agent, multi-path attacks and trade-offs between safety and operational continuity. To address this, our framework integrates hierarchical system modelling with Bayesian probabilistic reasoning, constructing Bayesian Network Graphs from system architecture and vulnerability data. Models are encoded using a Domain Specific Language to enhance computational efficiency and support dynamic updates. In our approach, we use a hybrid exposure probability estimation framework, which combines Exploit Prediction Scoring System and Common Vulnerability Scoring System scores via Bayesian confidence calibration to handle epistemic uncertainty caused by incomplete or heterogeneous vulnerability metadata. Mitigation recommendations are generated as countermeasure portfolios, refined using multi-objective optimisation to identify Pareto-optimal strategies balancing attack likelihood, impact severity, and system availability. To accommodate time- and resource-constrained incident response, frequency-based heuristics are applied to prioritise countermeasures across the optimised portfolios. The framework was evaluated through three representative cyber-physical attack scenarios, demonstrating its versatility in handling complex adversarial behaviours under real-time response constraints. The results affirm its utility in operational contexts and highlight the robustness of our proposed approach across diverse threat environments. 1 Introduction Cyber-Physical Systems (CPS) have become integral to daily life and infrastructure, enabling seamless interaction between digital and physical domains. Traditionally associated with Industrial Control Systems (ICSs), CPS are now extensively deployed in Connected and Autonomous Vehicles (CAVs), railway signalling infrastructures, and renewable energy networks. This expansion has drawn increased scrutiny toward cyber-physical security risks targeting critical infrastructure, from both attackers and defenders. In 2025, security researchers disclosed multiple vulnerabilities affecting solar photovoltaic (PV) inverters [ 10 ] , highlighting how CPS components could be exploited to induce service outages, cause infrastructure damage, or in severe cases, inflict physical harm. Given that many CPS deliver critical functions, real-time and adaptive decision support is essential for mitigating cyber incidents and minimising their impact. Here, “real-time” refers to the ability to assess and respond to threats within operational timeframes that preserve system safety and continuity, often under strict latency constraints. “Adaptive” refers to the capacity to dynamically adjust mitigation strategies based on evolving threat conditions, system states, and resource availability. However, several challenges remain. Unlike IT environments, where remediation typically involves deploying software patches or rebooting affected systems, CPS decisions must account for the interplay between cybersecurity and physical constraints [ 37 ] and take safety impacts into consideration. In legacy systems and safety-critical applications, patching a vulnerable asset introduces significant operational risks, as poorly tested patches may lead to system failures or degraded functionality due to hardware limitations. Decision-makers must also determine the scope and extent of intervention: how much to mitigate, which assets to secure, and where to prioritise responses. These decisions are further complicated by the presence of multiple attack paths and adversary tactics, necessitating an integrated approach to countermeasure selection that balances disruption mitigation with system availability. Recent decision-support systems (DSS) for CPS have introduced model-driven approaches leveraging simulation, system dynamics, and dependency modelling to guide risk mitigation planning [ 44 ] . While these systems support strategic assessment, they typically operate under static conditions and offer limited adaptability to evolving threats and system behaviours. For instance, most lack mechanisms to dynamically update model attributes, such as system properties or vulnerability exposure probabilities, during an ongoing cyber-physical incident. A typical example is the detection of a compromised communications dongle mid-response in a power grid, where the inability to update model parameters in real time reduces operational effectiveness in time-critical scenarios. To address these limitations, our research adopts probabilistic methodologies to enable real-time CPS decision support. We hypothesise that effective CPS incident response strategy must be guided by data-driven risk metrics and time constraints. To test this, our framework builds on: Bayesian Network (BN) models for dynamic risk management [ 33 ] ; threat modelling based on Failure Modes Effects and Criticality Analysis (FMECA) [ 3 ] ; a hybrid exposure probability estimation framework that synergistically integrates Common Vulnerability Scoring System (CVSS) [ 30 ] and Exploit Prediction Scoring System (EPSS) [ 21 ] scores through a novel Bayesian confidence calibration technique; the use of Domain-Specific Languages (DSLs) for semantic interoperability and computational efficiency [ 17 ] ; multi-objective optimisation, coupled with frequency-based heuristic analysis, to generate risk-informed countermeasure recommendations for decision-making [ 45 ] ; and finally, adaptive refinement of the model, informed by iterative assessments of observed countermeasure effectiveness. Figure 1: Architecture of our proposed framework Our proposed framework, as depicted in Figure 1 , is evaluated across three representative CPS scenarios: the 2015 Ukrainian power grid attack (BlackEnergy), a solar PV inverter system, and a railway Communication-Based Train Control (CBTC) system. Threat modelling for these scenarios was informed by the first author’s extensive cybersecurity experience in CPS, comprising 25 years of professional practice including 15 years as a Chief Information Security Officer. This depth of expertise ensured the inclusion of accurate technical insights and domain-relevant knowledge essential for credible CPS threat representation. Contributions . Our main contributions are: • We propose a new way to leverage Bayesian Networks for modelling cyber attack scenarios in CPS environments. Unlike previous methods, our models are enriched with system-specific and vulnerability metadata, enabling probabilistic reasoning and risk-informed recommendations for real-time decision support. To ensure structured representation and computational efficiency, the models are encoded in a DSL format, which also facilitates dynamic updates to support adaptive decision-making in response to evolving threats. • We use a hybrid exposure probability estimation framework that integrates the EPSS and the CVSS via a Bayesian confidence calibration technique. This framework yields robust and realistic recommendations under epistemic uncertainty stemming from incomplete or heterogeneous vulnerability metadata. • We propose a decision-support methodology for mitigating cyber incidents in CPS, centred on countermeasure portfolios derived through multi- objective optimisation and frequency-based heuristic analysis. As a novel contribution, mitigation probabilities are treated as parameters within a structured optimisation search space, enabling the generation of Pareto-optimal portfolios that form the foundation of the decision-support mechanism. frequency-based heuristic analysis across multiple optimisation runs is further employed to prioritise mitigation strategies under time and resource constraints. The methodology is evaluated across three representative CPS attack scenarios. • The models and source code developed in this research, along with reference data and formulae used in the framework, are publicly available via the project’s GitHub repository [ 18 ] . This supports transparency, reproducibility, and further development by both research and practitioner communities. The remainder of this paper is structured as follows. Section 2 provides background on the key techniques adopted in this work. Section 3 outlines the proposed methodology, including its theoretical foundations and implementation workflow, illustrated with a running example. Section 4 presents further applications of the framework to CPS attack scenarios, accompanied by an analysis of experimental results. Section 5 provides an in-depth, technical discussion of the findings and outlines directions for future research. Section 6 reviews related research and identifies limitations in existing approaches. Finally,