Data-driven Condition-based Maintenance Schedules of Active Distribution Networks

Scientific maintenance schedules are an effective way to reduce maintenance costs and promote asset management efficiency in distribution networks. With the increasing integration of a large number of stochastic distributed generators (DGs), the deterioration rate of distribution equipment may accelerate significantly owing to heavy/over loading. Conventionally, the distribution networks maintenance scheduling is based on run-to-failure or simplified scheduled maintenance strategies, without considering the integration of DGs and the health condition of distribution equipment. In this paper, the Markov decision model is applied to formulate the condition-based maintenance scheduling problem of active distribution networks. The operation of DGs and failure probability of distribution equipment are considered. The Q-learning method is adopted to optimize the maintenance schedules. Case studies show the effectiveness of the proposed method.