Reinforcement Learning for Optimal Allocation of Superconducting Fault Current Limiters

Superconducting fault current limiter (SFCL) is a new type of current limiting equipment that can restrict the short circuit current level in power systems without impacting on their normal operations. However, the existing SFCL allocation methods have disadvantages of long computation and unstable convergence. Recent advances in the area of artificial intelligence provide the possibility to achieve the optimal SFCL allocation with the consideration of both speed and convergence. To this end, the SFCL allocation is converted into a Reinforcement Learning (RL) problem in this paper. A designer to find the optimal SFCL allocation is viewed as an agent, and the power system is regarded as the environment. The agent adjusts the location and size of SFCLs to respond to the feedback from the environment optimally. The Q-learning algorithm is applied to solve the RL problem. The experimental results demonstrate the effectiveness and superiority of the proposed method.