Brain-Inspired Deep Reinforcement Learning for Load Frequency Control of Power Systems With Composite Interference

This article proposes a brain-inspired deep reinforcement learning (DRL)-based load frequency control framework for power systems with composite interference of internal noise and external load fluctuation. Specifically, inspired by the decision-making process of human brain, some historical, present, and future characteristics of system states are thoroughly extracted into the experience pool for the DRL agent to train efficiently. Meanwhile, a progressive training mechanism that divides the training process into stages through gradually increasing the training objectives is designed for less blindness during training. Moreover, in response to the composite interference, some simulative interferences are learned beforehand to promote the adaptability of the DRL agent. Experimental results on a power system of single generating unit demonstrate that the proposed method is competent in controlling the frequency while ensuring the load supply under composite interference, and is superior to the comparative methods in control performance and training efficiency.