Distributed Post-Disaster Restoration of Interdependent Infrastructure Networks under Spatially Correlated Regional Hazards

Regional hazards can severely disrupt interdependent infrastructure networks, and post-disaster restoration is further complicated by spatially correlated repair uncertainty and limited data sharing among networks. This paper proposes a distributed post-disaster restoration framework for interdependent power and water networks with a known initial damage set. Spatially correlated repair durations are generated through a Gaussian copula, and a representative scenario set is generated via a scenario reduction method. A two-stage stochastic mixed-integer model determines a repair schedule in Stage 1, while Stage 2 optimizes network operation under each repair-duration scenario. To coordinate different operators under privacy constraints an Alternating Direction Method of Multipliers with embedded Logic-Based Benders Decomposition (ADMM-LBBD) is developed and only the status of dependency-pair nodes is exchanged through shared gate variables in this method. Applied to the Shelby County power-water system, the distributed method converges in 9 ADMM iterations and yields a cumulative satisfied demand of 3635, compared with 561 for the no-repair benchmark and 3994 for centralized coordination. The solution also reveals a consistent power-first, water-follows recovery pattern induced by physical interdependencies. Overall, the proposed method achieves performance within 91 % of a centralized solution while preserving data privacy and maintaining tractable computation.

Keywords: Interdependent infrastructure networks, decentralized post-disaster restoration, two-stage stochastic programming, spatial correlation, ADMM-LBBD.