The growing reliance on digital communication and automation in modern distribution networks has created new vulnerabilities to coordinated cyber attacks. A particular threat—stealthy false data injection (S-FDI) attacks—can bypass conventional bad data detectors and propagate through physical tie-line couplings that interconnect multiple microgrids, potentially triggering cascading failures across entire clusters.
Researchers have introduced BR-FedMAPPO, a Byzantine-resilient federated multi-agent reinforcement learning framework designed specifically for secure operation of interconnected microgrids. The system allows multiple microgrids to cooperatively optimize their dispatch and control strategies while remaining resilient to malicious attacks and protecting sensitive operational information.
The framework operates by deploying local agent-critic learning algorithms at each microgrid. Critically, the learning models are structured so that sensitive configuration data—such as locations and ratings of distributed flexible AC transmission (D-FACTS) devices, battery energy storage systems, and inter-microgrid tie-line capacities—remain private at each site. Only high-level operational policies are shared across the federated network.
The defense mechanism operates in multiple layers. First, a Byzantine-resilient aggregation algorithm filters suspicious updates using trimmed-mean methods and reward-weighted combinations to reject malicious model manipulations. Second, each microgrid learns a "moving target defense" strategy that continuously adjusts reactive power control, energy storage dispatch, and inter-microgrid power exchanges to maintain operation despite attacks. Third, the system includes an adaptive islanding capability that can automatically isolate compromised connections when necessary.
Validation on interconnected test systems based on IEEE 30-bus and 118-bus networks demonstrated effective attack detection and containment of cascading disruptions. The framework maintained economical dispatch performance while simultaneously protecting grid stability, operational secrecy, and learning channel integrity against coordinated false data injection attacks.
This work addresses a critical gap in microgrid security by enabling collaborative optimization across multiple independent operators without exposing proprietary infrastructure details or creating centralized cybersecurity vulnerabilities.
