School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798, Singapore

While the flexibility of power system operation can be improved by topology optimization, the dimension of system-level discrete decision variables, including the connections of lines and substation busbars, is prohibitively high. Thus, the topology optimization problem of power systems can hardly be solved by the conventional mixed-integer optimization method. Aiming at this problem, a reinforcement learning based method is proposed combining asynchronous advantage Actor-Critic (A3C) and power system domain knowledge, which transfers the computational burden of online optimization to the offline agent training stage. The defined reward function is adopted to minimize the violations of power transmission line flow limits. Forced constraints verification is employed to reduce the searching space and improve the efficiency of the reinforcement learning. The fast computation of the topology optimization of power system operation is realized,and the operation security of power systems is enhanced. The effectiveness of the proposed method is validated by simulation testing results.