Resilience-based post disaster recovery optimization for infrastructure system via deep reinforcement learning

• A resilience-based DRL sequential decision-making framework is developed for optimizing infrastructure system recovery, leveraging a graph-based representation of the network topology and proposing the LoR to lead the optimization. • A worst-case scenario training strategy is proposed for the DRL-based model to improve its adaptability and robustness in unseen disaster scenarios. • A fair comparison of different DRL architectures is provided, demonstrating that DDQN outperforms other algorithms on this recovery optimization task. • Superior effectiveness and computational efficiency in optimizing recovery strategies were demonstrated by the DRL approach over genetic algorithms. Infrastructure systems are essential yet vulnerable to natural and man-made disasters. Efficient post-disaster recovery requires repair-scheduling approaches under limited resources shared across the system. Existing approaches like component ranking, greedy algorithms, and data-driven models often lack resilience orientation, adaptability, and require high computational resources when tested within such a context. To tackle these issues, we propose a solution by leveraging Deep Reinforcement Learning (DRL) methods and a specialized resilience metric to lead the recovery optimization. The system topology is represented adopting a graph-based structure, where the system’s recovery process is formulated as a sequential decision-making problem. Deep Q-learning algorithms are employed to learn optimal recovery strategies by mapping system states to specific actions, i.e., which component ought to be repaired next, with the goal of maximizing long-term recovery from a resilience-oriented perspective. To demonstrate the efficacy of our proposed approach, we implement this scheme on the example of post-earthquake recovery optimization for an electrical substation system. A comparative analysis against baseline methods reveals the superior performance of the proposed method in terms of both optimization effect and computational cost, rendering this an attractive approach in the context of resilience enhancement and rapid response and recovery.