Enhancing Security of Distributed Multi-Agent Systems in Smart Grids: An AI-Driven Approach to Regular Audits

Smart grids are increasingly susceptible to cyber-physical anomalies due to their complex interconnections and reliance on real-time data. Ensuring grid reliability and minimizing operational risks require robust anomaly detection systems and cost-effective audit strategies. This paper proposes a novel multi-objective framework that integrates anomaly detection with dynamic audit optimization to address these challenges. The framework leverages real-time operational data ( \(X\) ), baseline metrics ( \(B\) ), thresholds ( \(Th\) ), and criticality weights ( \(w\) ) to identify anomalies and dynamically assess system risks. Using a deviation-based scoring mechanism, anomalies are detected across critical grid components such as power generators, breakers, and controllers, while audit frequencies are optimized to balance attack rates and operational costs. In addition, this research contributes to Embodied AI by integrating intelligent audit agents within a physical smart grid environment; representing perception-action cycles that monitor, learn, and act in response to environmental conditions which is an essential characteristic of Embodied AI systems. These agents learn from real-time interactions and adapt behavior dynamically, aligning with the embodied AI paradigm. The framework's multi-objective optimization model minimizes the attack rate ( \(R_{attack}\) ) and the total cost ( \(C\) ) by dynamically adjusting audit schedules based on real-time risk trends. The proposed solution is validated using simulated scenarios of cyberattacks (e.g., false data injection) and physical faults (e.g., overcurrent, voltage sag). Comparative analysis with traditional fixed-audit schedules demonstrates significant improvements, including 42.5% cost efficiency, 93.8% audit coverage, and 98.4% anomaly detection accuracy. The results establish the framework as a scalable and adaptive solution for anomaly management in smart grids, advancing the field of secure and efficient grid operations. This work contributes to the development of intelligent, risk-aware methodologies for grid resilience and operational optimization. The findings lay the groundwork for further exploration of AI-enhanced security solutions in autonomous and distributed systems.