Resilient Load Restoration With Distributed Mobile Resources: A Model-Free Approach

High-impact, low-probability natural disasters can lead to significant network failures, which would compromise the resilience of the power system. Distributed mobile resources (DMRs), such as mobile energy storage systems, heavy-duty electric trucks, and private electric vehicles, can enhance power system resilience through flexible charging and discharging scheduling. In this paper, we propose a deep reinforcement learning based method for efficiently dispatching DMRs to rapidly restore critical loads following disasters. Initially, we formulate DMRs' routing and scheduling problem as a Markov decision process. Subsequently, we propose a detailed physical hurricane model, aiming to calculate equivalent wind speeds for specific segments of distribution lines and further estimate the likelihood of line damage. Lastly, the double deep Q-network method is developed to address the load recovery issue, wherein DMRs are treated as agent that generate real-time routing and discharging strategies contingent on damage assessments of current system operational states. Our proposed model-free algorithm exhibits superior performance in characterizing disaster scenarios and accelerates the decision-making of optimal scheduling strategies. Case studies conducted on 6 and 33 bus systems demonstrate that DMRs can collaborate effectively, significantly enhancing the system's resilience in disaster situations.