Coordinated load frequency control of multi-area integrated energy system using multi-agent deep reinforcement learning

To dynamically balance multiple energy fluctuations in a multi-area integrated energy system (IES), a coordinated power control framework, named distributed intelligent coordinated automatic generation control (DICAGC), is constructed among different areas during load frequency control (LFC). Furthermore, an evolutionary imitation curriculum multi-agent deep deterministic policy gradient (EIC-MADDPG) algorithm is proposed as a novel deep reinforcement learning algorithm to realize coordinated control and improve the performance of DICAGC in the performance-based frequency regulation market. EIC-MADDPG, which combines imitation learning and curriculum learning, can adaptively derive the optimal coordinated control strategies for multiple areas of LFC controllers through centralized learning and decentralized implementation. The simulation of a four-area LFC-IES model on the China Southern Grid (CSG) demonstrates the effectiveness of the proposed method in maximizing control performance while minimizing regulation mileage payment in every area against stochastic load and renewable power fluctuations.

Automated summary

A coordinated power control framework and a novel deep reinforcement learning algorithm EIC-MADDPG are proposed to achieve coordinated control and improve performance in a multi-area integrated energy system (IES). By combining imitation learning and curriculum learning, the algorithm can adaptively derive optimal coordinated control strategies for multiple LFC controllers.