Maintenance Decision Generator for Electrical Equipment Based on Reinforcement Learning

The maintenance of electrical equipment aims to guarantee a good working condition of the equipment and to increase the stability of the power grid operation. In this paper, we propose a novel maintenance decision model for both single and multiple electrical equipment, based on the Markov hypothesis of equipment states and reinforcement learning. Specifically, the cut set of the power grid is incorporated to calculate the weights of different equipment in multiple equipment mode, which are then applied to the dynamic programming solution to make the learned strategies focus on the differences between equipment. Moreover, the current cut set is used to recalculate action rewards to indirectly implement the communication between action sequences with the knowledge of that maintenance actions can lead to the changes of cut set. Experimental results demonstrate that the proposed method significantly improved the effectiveness of decision making.