Multi-Agent Distributed Deep Learning Algorithm to Detect Cyber-Attacks in Distance Relays

Distance relays are critical components in protection systems of power grids that can be attacked by cyber-attackers. Indeed, a cyber-attacker injects fake data into distance relay to pretend fault has happened, and the must tripped. Thus, new powerful approach, named Multi-Agent Distributed Deep Learning (MADDL) method is proposed tackle cyber-attacks relays. Unlike centralized methods, system with several mapped multi-agent distributed employing graph theory, which considered as agents or nodes graph. Each agent only connected neighboring exchange voltage current data. Then, deep neural network cyber-attack detection structure assumed for each utilizes local received from detect attacks. Hence, structures tuned train data, obtained simulating grid different types faults. evaluated test dataset, including under various faults normal situation injecting cyber-attacks. The developed been employed three case studies, IEEE 6-bus, 14-bus, 118-bus grids. According simulation results, algorithm succeeded identifying more than 99.88%