Optimal Maintenance Scheduling of Transmission Systems with a Reinforcement Learning Approach

Timely detection of potential risks in regional power grid equipment can be achieved through rational planning of daily maintenance schedules, which holds paramount significance for protecting regional power supply. Therefore, this paper introduces an optimization method for power grid maintenance schedule based on Q-value. Firstly, we formulate a Markov optimization model for scheduling maintenance, aiming to maximize power grid security and stability, minimize power quality issues, and optimize execution utility, all integrated within a weighted objective function. Secondly, we propose a new planning method for maintenance schedule state space, leveraging the epsilon-greedy action selection under the constraint of schedule continuity. It is then applied to the Q-learning state judgment to optimize the maintenance scheduling. Finally, the proposed method is applied to the IEEE-39-bus system for scheduling maintenance tasks, and experimental results demonstrate its feasibility and rationality.