Strategic Management and Seismic Resilience Enhancement of Water Distribution Network Using Artificial Neural Network Model

The seismic resilience of water distribution networks (WDNs) as the most critical urban infrastructure is a major aspect of crisis management. Developing accurate computational models to evaluate the seismic behavior of WDNs and enhance seismic resilience remains a major research challenge. This paper introduces a novel model to evaluate the seismic vulnerability of WDNs through an artificial neural network (ANN) approach. To generate initial data to develop the seismic damage prediction model, extensive numerical modeling was performed using commercially available software to extract the strain behavior of buried pipes under seismic loading. A total of 720 numerical simulations were performed. The numerical model was validated using an experimental model. The results showed that the numerical model had an error smaller than 10% in evaluating the strain behavior of buried pipes, suggesting satisfactory model performance. This paper used a multilayer perceptron (MLP) and the Levenberg–Marquardt algorithm (LMA) because it has shown good performance in regression applications. The sensitivity of the proposed model to the number of hidden layers was analyzed, and the MLP with 15 hidden layers was found to be optimal in predicting the strain of the buried pipe under seismic loading, with a mean squared error (MSE) of 0.301 and a correlation coefficient R=0.969. The proposed seismic damage prediction model was executed on the WDN of Tehran, Iran, based on the initial data set under five seismic loading scenarios, calculating the numbers of breaks and leaks. The seismic resilience of the WDN was evaluated using damage, minimum water demand, and restoration time indices. Several strategies were proposed to enhance the seismic resilience of WDNs. The developed seismic resilience assessment model in this study has the capability to be applied and implemented across other WDNs.