A deep reinforcement learning-based approach for cyber resilient demand response optimization

Abstract

The contemporary smart grid infrastructure, characterized by its bidirectional communication capabilities between prosumers and utility organizations, has revolutionized the efficient execution of fine-grain computational tasks. Ensuring the uninterrupted delivery of power, even in the face of unforeseen contingencies, stands as a paramount concern for utility companies. Peak load forecasting, load balancing, and robust cyberattack detection and prevention mechanisms are integral components in achieving grid reliability. This research endeavors to advance peak load forecasting strategies and demand response optimization at the microgrid level, thereby enhancing grid reliability through the application of Deep Reinforcement Learning (DRL) techniques. Additionally, it investigates the ongoing threat of false data injection attacks. By synergizing these two critical investigations and implementing a novel framework and defense mechanism, this paper proposes a comprehensive approach to fortify the smart grid’s reliability and security. The envisioned framework not only refines demand response (DR) optimization but also bolsters the grid’s resilience in the face of the everevolving cyber threat landscape. The research outcomes showcase the practicality and effectiveness of the proposed framework, substantiated through extensive experimentation conducted on IEEE-3, IEEE-9, IEEE-14, and IEEE-33 bus systems.