Counterfactual Multi-Agent Policy Gradient-Based LFC Method for Interconnected Power System

Load frequency control (LFC) is crucial for maintaining the stability of interconnected power system. However, the complex dynamic couplings in multi-area systems and the lack of an effective credit assignment mechanism in traditional methods limit their coordination and control performance. To address multi-agent coordination challenges in power systems, this work presents an innovative LFC methodology grounded in the counterfactual multi-agent (COMA) policy gradient algorithm. The method employs a centralized critic and counterfactual baselines to optimize credit assignment and utilizes an advantage function to evaluate agent behaviors, thereby optimizing generation commands. In this framework, each control area is treated as an independent agent that dynamically adjusts generation output based on local observations and global feedback, interacting with the interconnected power system. The COMA algorithm enables each agent to assess its contribution to system stability, ensuring cooperative frequency regulation. Simulation results on a three-area power system demonstrate the effectiveness of the proposed method, improving LFC performance and system robustness.