Water distribution network risk assessment and AI-aided rapid recovery under seismic hazard

Water distribution networks (WDNs) are critical infrastructures that provide essential support for community life. The service of WDNs can be significantly interrupted under seismic hazards due to a large number of failures. The resilience of a WDN describes its ability to withstand the seismic as well as recovery from damage situations. This paper introduces a framework that simulates the failure and recovery process of WDN under the seismic damage and a machine learning (ML) model for resilient and rapid decision-making after the hazard occurs. The simulation framework considered the water pipe location, WDN graph structure, pipe physical attributes, and dynamic customer water consumption. Based on the proposed simulation model, a reinforcement learning (RL) model is trained to obtain resilient decision-making after damage from hazard occurs. The results show that a reliable risk assessment and recovery simulation framework is the key for training a RL model. With the trained RL model, the optimal decisions post-hazard can be determined rapidly even the model has never seen this hazard before. This study demonstrates the potentials of artificial intelligence (AI) techniques to support the optimal decision in managing the WDN under emergency conditions.