Deep Learning of Complex Pipe Leakages Events in Drinking Water Distribution Networks for Effective Spatiotemporal Pre-Detections and Isolations of Leak Conditions

To detect pipes leakages over space and time of the water distribution in L-TOWN, this research study develops an alternative engineering tool, by combining the numerical capabilities of genetic algorithm and deep learning, which can pre-detect near and/or exact locations of pipe leakages within the water distribution network in L-TOWN over time. The genetic algorithm is programmed using an open-source Water Network Tool for Resilience (WNTR) in Python package. WNTR is an EPANET compatible Python version and is designed to simulate and analyze resilience of water distribution networks. For the deep learning component, a personalized feed-forward deep neural network (DNN) is built on Tensorflow platform to develop a trained predictive model using volumes of calibrated simulation data derived from WNTR based on the physical characteristics of the water distribution network in L-TOWN. The trained DNN model is then leveraged to predict the near and/or exact locations of pipe leakages in L-TOWN using the real-world measured data from the reported years of 2018 and 2019.