Multi-agent reinforcement learning design of load-frequency Control with frequency bias estimation

Conventional load-frequency control (LFC) systems use proportional-integral (PI) controllers. These controllers are designed based on a linear model and the nonlinearities of the system are not accounted for. Then they are incapable to gain good dynamical performance for a wide range of operating conditions. A control strategy for solving this problem in a multi-area power system is presented by using a multi-agent reinforcement learning (MARL) approach based on the frequency bias (β) estimation that genetic algorithm (GA) optimization is used to tune its parameters. This approach contains two agents in each control area, estimator agent and controller agent that communicate with each other. The proposed method does not depend on any knowledge of the system and finding area control error (ACE) signal based on the frequency biased estimation, improves the LFC performance. To demonstrate the capability of the proposed control structure, a three-control area power system simulation with two different scenarios is presented.