Research on Fast Optimal Corrective Transmission Switching Based on Deep Reinforcement Learning

Industrially, overloads are required to be eliminated within 15 minutes. However, with the increasing proportion of uncertain renewable energy resources, it is difficult to analyze each scenario online, which poses potential risk for power system operation. Addressed to the real-time overload elimination demand, an optimal Corrective Transmission Switching (CTS) framework is constructed based on deep reinforcement learning. First, the overload amount is deduced based on power transfer distribution factors (PTDF) to accelerate the correction procedure. Then, considering the instant response of CTS, compared with generation rescheduling and load shedding, we formulate an optimal CTS model accordingly for further speed up. Finally, taking the advantage of real time interaction with the environment, we apply the deep reinforcement learning for model solving, and a two-stage A3C Optimal CTS framework is built up to generate CTS strategies for overload elimination. The effectiveness and rapidity of the proposed method are verified by the modified IEEE 118-bus system.