A COMBINED DEEP REINFORCEMENT AND SUPERVISED LEARNING TECHNIQUE TO IDENTIFY CYBER-ATTACKS IN DISTANCE RELAYS

Abstract

This paper proposes a Multi Agent Deep-reinforcement learning algorithm for detecting cyber-attacks in distance relays. Distance relays are widely used in the power grid for protection against faults, but they are vulnerable to cyber-attacks due to their distributed nature. The proposed algorithm uses multiple agents to learn an optimal policy for detecting cyber-attacks in distance relays. Each agent is trained using a combination of deep reinforcement learning and supervised learning techniques. The agents are trained to identify attacks by observing the current state of the system and taking actions that optimize a reward function. The reward function is designed to maximize the detection accuracy of the agents while minimizing the false alarm rate. The algorithm is evaluated on a benchmark dataset of simulated cyber-attacks. The results show that the proposed algorithm outperforms existing approaches in terms of attack detection accuracy and false alarm rate.