Reconstruction of Load Response Behavior Based on Multi-Agent Generative Adversarial Imitation Learning

This paper explores the use of Multi-Agent Generative Adversarial Imitation Learning (MAGAIL) to model and reconstruct electricity consumer behavior in demand response (DR). DR is a crucial tool for balancing supply and demand in the power system, reducing peak-hour demand, and enhancing grid reliability and flexibility. Traditional consumer behavior reconstructing models struggle with high-dimensional, nonlinear, dynamic, and highly personalized challenges. The GAIL-based method simulates expert behavior strategies and learns from historical consumer data without the need for explicit reward functions, generating strategy models that closely resemble real consumers' behavior. This paper introduces the application of GAIL in load response reconstruction and validates the model's effectiveness using real-world data from the Pennsylvania-New Jersey-Maryland (PJM) electricity market.