Innovative framework for cost-effective and reliable networks via P-graph and machine learning in water infrastructure management

Abstract

Abstract Ensuring the reliable operation of wastewater transportation networks is critical for maintaining urban water infrastructure performance and preventing service disruptions. However, reliability assessment is typically conducted as a post-design evaluation rather than being explicitly integrated into the network synthesis and optimization process. This study proposes an integrated framework that combines Machine Learning (ML)-based reliability estimation with Graph-theory-based process optimization (P-graph) to support cost-effective infrastructure planning and retrofit decision-making. XGBoost and Neural Network models were developed to estimate pipeline failure rates using real data on physical, operational, and contextual characteristics of wastewater pipelines. The predicted failure rates were subsequently converted into reliability metrics and incorporated into a P-graph optimization framework to evaluate alternative retrofit configurations. The proposed methodology was demonstrated using a wastewater transportation network case study derived from the LeakDB benchmark dataset. The ML models effectively identified pipeline segments with elevated failure risk, providing valuable insights for asset management and maintenance prioritization. The optimization framework generated Pareto-optimal retrofit strategies that balance reliability enhancement and economic investment. Results indicate that the inclusion of distributed storage tanks and redundant pipelines substantially improves system reliability, achieving reliability levels exceeding 0.950 under full storage capacity. Sensitivity analysis further reveals that storage capacity is a critical determinant of system reliability, with significant reductions observed when the capacity falls below 50% of the required volume. The proposed framework offers a systematic approach for integrating data-driven reliability prediction with infrastructure optimization, enabling more reliable and cost-effective wastewater infrastructure management.