Uncertainty-Inflicted Event-Driven Resilient Recovery for Distribution Systems: A Semi-Markov Decision Process Approach

Repair and reconfiguration are vital for power recovery after outages caused by natural disasters in distribution systems, but sequential and uncertainty-inflicted decision points due to uncertain repair periods make power recovery complicated. This paper proposes semi-Markov decision process (SMDP)-based resilient recovery with sequentially event-driven repair and reconfiguration in consideration of uncertainty-inflicted decision-making points. The sequential repair/reconfiguration actions in consideration of uncertain repair periods are considered as uncertainty-inflicted event-driven processes. The sequential repair states with different repair crews are established as semi-Markov states. The whole sequential and uncertain decision-making process is modeled as a semi-Markov decision process-based optimization model, which is an event-driven recursive model. Q-learning is employed to solve the proposed model, and the convergent estimations of Q values for semi-Markov states map the original model into an event-driven deterministic optimization based on the sequential repairs that actually occurred over the time horizon. IEEE 123-bus system is used to validate the proposed model.