Strategic Electrical Asset Maintenance by Reinforcement Learning

With the penetration of renewable energy sources (RESs), the operation of power system is becoming more challenging. For stable operation of renewable energy-rich power system, electrical assets need to switch to different operational conditions more frequently, resulting in accelerated aging problems. To reduce downtime cost resulted from failure of electrical assets, an appropriate maintenance strategy is crucial. Traditionally, time-based preventive maintenance is conducted to manage the lifecycle of electrical assets, but this maintenance strategy is inefficient and costly. To improve operation efficiency and reduce downtime cost, condition-based maintenance (CBM) is proposed. While CBM can use the insitu operating data to assess the health of electrical assets for better lifecycle management, the determination of optimal CBM requires expert knowledge, which is not applicable to new electrical assets. This paper proposes a framework to determine optimal CBM by utilizing the self-exploration and self-learning capabilities of reinforcement learning. There are two steps to establish this intelligent maintenance strategy: The first step is to model a dynamic environment which can be updated with real-time operating data, and the second step is to obtain the optimal maintenance strategy based on reinforcement learning algorithm. The framework proposed in this paper can be applied to all types of electrical assets including old and new assets, overcoming the drawbacks of the expert knowledge-based maintenance strategy planning.