Routing and Scheduling of Mobile Energy Storage Systems for Distribution Network Resilience Enhancement Based on a Hybrid Data-Model Driven Approach

Mobile Energy Storage Systems (MESSs) are critical for improving distribution network resilience under extreme weather events. However, the mathematical model for MESS routing and scheduling is essentially a high-dimensional mixed-integer nonlinear stochastic optimization problem. To accurately and rapidly obtain the routing and scheduling of MESSs and the optimal operation of distribution networks, a bi-level optimization of MESS routing and scheduling based on a hybrid data-model driven approach is proposed. In the upper level, the data-driven approach is achieved by the graph attention network multi-agent conservative soft actor-critic reinforcement learning (GMARL) to determine optimal routing decisions for MESS in the transportation network while taking into account traffic flow and road repair time uncertainties. The proposed GMARL takes full advantage of a graph attention network for feature extraction, and adopts the multi-agent conservative soft actor-critic to mitigate overestimation caused by out-of-distribution experiences, thereby effectively coordinating multiple MESSs to achieve the optimal strategy. In the lower level, a mixed-integer second-order cone programming is formulated to obtain optimal scheduling strategies for MESSs, reconfiguration and optimal power flow in the distribution network. Upon determining the scheduling strategies and amount of load recovery, the reward function value for each MESS can be calculated and used to update the neural network parameters of GMARL, thereby further optimizing the routing strategies of MESSs. Finally, case studies on an IEEE 33-bus active distribution network and a 12-node transportation network are conducted to verify the effectiveness of the proposed approach.