Probabilistic protection of smart power systems against load redistribution attacks using ai-based critical measurement selection

Abstract

The modernization of smart power systems through the integration of artificial intelligence (AI), Industrial Internet of Things (IIoT), cloud computing, renewable energy resources, and advanced communication infrastructures has significantly improved operational efficiency and grid intelligence. However, the increased interconnectivity between Information Technology (IT) and Operational Technology (OT) environments has simultaneously exposed smart grids to sophisticated cyber threats. Among these threats, Load Redistribution Attacks (LRAs), a critical class of False Data Injection Attacks (FDIAs), can manipulate system measurements while approximately preserving the reported system load, thereby bypassing traditional bad-data detection methods and causing severe operational and economic impacts. This manuscript proposes a probabilistic protection framework against LRAs using AI-based critical measurement selection for resilient smart power system operation. Unlike prior single-mechanism defenses, the proposed framework tightly couples an adaptive, sliding-window AI anomaly detector, a quantitatively defined probabilistic cyber-risk index, a machine-learning-based critical measurement selection engine, a formally specified zero-sum game-theoretic protection allocator, and a Zero Trust Architecture (ZTA) with IEC 62443-aligned OT segmentation into a single closed-loop defense pipeline, so that each stage adapts using the outputs of the others. Mathematical formulations for the DC-approximated attack constraint (with a discussion of its extension to AC power flow), cyber risk, attack probability, vulnerability, operational impact, and economic loss are developed and validated on the modified IEEE 14-, 30-, and 57-bus systems under multiple operating and attack-intensity conditions. Simulation results, including confusion-matrix-based metrics, ROC analysis, and comparison against conventional IDS and rule-based baselines, indicate that the proposed framework improves attack detection accuracy, reduces economic losses, and enhances system resilience relative to conventional deterministic protection approaches. Limitations of the DC-approximation-based validation and directions for AC state-estimation-based and graph-neural-network-based extensions are discussed.