Grid-area coordinated load frequency control strategy using large-scale multi-agent deep reinforcement learning
Jiawen Li, Jian Geng, Tao Yu
Abstract
In order to enable full participation of high-performance units controlled by different dispatching centers in the performance-based frequency regulation market, a data-driven grid-area coordinated load frequency control (GAC-LFC) strategy using unified performance-based frequency regulation market mechanism is proposed. The strategy takes into account the coordination of LFC controllers in different areas of the interconnected power grid, and accommodates a large number of high-performance units controlled by dispatching centers in secondary frequency regulation. In addition, an effective exploration-based multi-agent deep deterministic policy gradient (EE-MADDPG) algorithm is proposed as the framework algorithm. In this algorithm, the LFC controller controlled by the grid-dispatching center and the LFC controller controlled by the area-dispatching center in each area are treated as different agents. Through centralized training with decentralized execution, the coordination of LFC controllers controlled by different levels of dispatching centers in different areas can be realized. Moreover, the algorithm introduces effective exploration strategies, agents operating on various principles, and artificial intelligence functions based on imitation learning and curriculum learning, which altogether constitute a more robust strategy. Through the simulation of the four-area LFC model of the China Southern Grid (CSG), it is demonstrated that the proposed method can simultaneously call more high-performance units, improve multi-area LFC control performance, and reduce the frequency regulation mileage payment in each area.