A distributed deep reinforcement learning-based approach for fast preventive control considering transient stability constraints

Preventive transient stability control is an effective measure for the power system to withstand high-probability severe contingencies. It is mathematically an optimal power flow problem with transient stability constraints. Due to the constraints involved the differential algebraic equations of transient stability, it is difficult and time-consuming to solve this problem. To address these issues, this paper presents a novel deep reinforcement learning (DRL) framework for preventive transient stability control of power systems. Distributed deep deterministic policy gradient is utilized to train a DRL agent that can learn its control policy through massive interactions with a grid simulator. Once properly trained, the DRL agent can instantaneously provide effective strategies to adjust the system to a safe operating point with a near-optimal operation cost. The effectiveness of the proposed method is verified through numerical experiments conducted on New England 39-bus system and NPCC 140-bus system.