Securing Smart Grids: Deep Reinforcement Learning Approach for Detecting Cyber-Attacks

Smart meters (SMs) are deployed in smart power grids to monitor customer power consumption and facilitate energy management. However, fraudulent customers can compromise these SMs to manipulate power readings and engage in electricity theft cyber-attacks, resulting in reduced electricity bills. While various machine learning approaches have been employed for detecting such attacks, the potential of reinforcement learning (RL) remains unexplored. To bridge this gap, we propose a deep reinforcement learning (DRL) approach that leverages RL's adaptability to dynamic cyber-attacks and consumption patterns. This approach integrates exploration and exploitation mechanisms, enabling optimal decision-making. In this study, we present our approach in two scenarios. Firstly, we develop comprehensive detection models using deep Q networks (DQN) and double deep Q networks (DDQN) with various deep neural network architectures. Secondly, we address the challenges of defending against newly launched cyber-attacks. Extensive experimentation provides strong evidence of the effectiveness of our DRL approach in improving the detection of electricity theft cyber-attacks, as well as its capacity to efficiently adapt and defend against newly launched cyber-attacks.