EPyT-Flow: A Toolkit for Generating Water Distribution Network Data

Abstract

This work introduces EPyT-Flow, an open-source Python package building on top of EPyT for facilitating water distribution network (WDN) simulations. EPyT-Flow provides a high-level interface for the easy generation of hydraulic and water quality scenario data. Additionally, it provides access to low-level functions of EPANET and EPANET-MSX for hydraulic and water quality modeling respectively. To accelerate research in WDN management, EPyT-Flow provides easy access to popular benchmark data sets for event detection and localization, and an environment for developing and testing feedback control algorithms. Statement of need Water Distribution Networks (WDNs) are designed to ensure a reliable supply of drinking water. These systems are operated and monitored by humans, supported by software tools, including basic control algorithms and event detectors that rely on a limited number of sensors within the WDN. These sensors measure hydraulic (e.g., pressure, flow) and water quality (e.g., chemical concentrations) states. However, given the rapid population growth of urban areas, WDNs are becoming more complex to manage due to the resulting time-varying system uncertainty. Consequently, key tasks such as event detection (e.g., leakage) and isolation, pump scheduling, and control are becoming more challenging. Moreover, modeling and predicting water quality in the distribution network is becoming more difficult due to changing environmental conditions. This is why water utilities are now driven to install even more sensors to gather data on their changing systems. Traditionally, model-based methods were used for planning and managing WDNs; however, due to rapid changes, these methods may no longer be sufficient. New AI and data-driven methods can now take advantage of big data and are promising tools for tackling challenges in water management. Currently, non-water experts such as AI researchers face several challenges when devising practical solutions for water system applications, such as the unavailability of tools for easy scenario/data generation and easy access to benchmarks, which hinder the progress of applying AI to this domain. Easy-to-use toolboxes and access to benchmark data sets are extremely important for boosting and accelerating research, as well as for supporting reproducible research. This was, for instance, the case in deep learning and machine learning with toolboxes such as TensorFlow and scikit-learn.