Reinforcement Learning Based Real-Time Wide-Area Stabilizing Control Agents to Enhance Power System Stability

In this paper, the design of a network of real-time close-loop wide-area decentralized power system stabilizers (WD-PSSs) is investigated. In this approach, real-time wide-area measurement data are processed and utilized to design a set of stability agents based on a Reinforcement Learning (RL) method. Recent technological breakthroughs in wide-area measurement system (WAMS) make the use of the system-wide signals possible in designing power system controllers. The main design objectives of these controllers are to stabilize the system after severe disturbances and mitigate the oscillations afterward. The proposed stability agents are decentralized and autonomous. The proposed method extends the stability boundary of the system and achieves the above goals without losing any generator or load area and without any knowledge of the disturbances causing the response. This paper describes the developed framework and addresses different challenges in designing such a network. A case study is provided to illustrate and verify the performance and robustness of the proposed approach.