Attributing minimum night flow to individual pipes in real-world water distribution networks using machine learning

Abstract

This article introduces an explainable machine learning model for estimating the amount of flow that each pipe in a district metered area (DMA) contributes to the minimum night flow (MNF). This approach is validated using the MNF of DMAs and pipe failures, showing good results for both tasks. The predictions from this model could be used to guide leak management or intervention strategies. In total, 800 DMAs ranging from rural to urban networks and representing nearly 12 million meters of pipe from a UK water company are used to train, validate, test, and evaluate the methodology.