Real-Time Service Restoration of Coupled Power-Water Systems Considering the Spatio-Temporal Cascading Failure

The interdependence between power distribution systems (PDS) and water distribution systems (WDS) plays a critical role in the coupled power-water systems (CPWSs) service restoration. However, the transient water flow dynamics in WDS were omitted in existing studies, leading to impractical restoration plans. To address this gap, this paper proposes a decision-making method for the CPWSs service restoration considering the spatiotemporal cascading failure. By integrating a WDS transient hydraulic model into the CPWSs service restoration optimization, this method enables dynamic adjustments of repair crew dispatch and network reconfiguration plans to adapt to water flow dynamics. Moreover, to enable real-time and adaptive decisionmaking for unexpected contingencies in extreme events, an offline deep reinforcement learning (DRL) approach is introduced. By pre-training on a dataset collected from the proposed stochastic contingencies simulations, this method can remarkably mitigate the computational burden associated with WDS transient flow analysis, and eliminate the need for frequent agent–environment interactions. Case studies on a modified 14-bus/15-node CPWS and a 118-bus/71-node CPWS are conducted to demonstrate the effectiveness and robustness of the proposed method.