Identification of High Pressure Critical Links in Water Distribution Systems

Control system is the core of the water distribution system (WDS). WDS system has information in the form of data feeds. By analyzing these data feeds, we can provide enough information about the leakage area in the system to take control manually over the system. In this paper, leakage areas are isolated using various Machine Learning (ML) techniques. EPANET (Environmental Protection Agency Networks) is used to model the hydraulic behavior of WDS. In WDS, leakage scenarios are created through EPANET by varying the input parameters (such as roughness, tank level, base demand). This data is then analyzed using different supervised machine learning techniques to determine the high-pressure links to identify the critical areas. We also compare the accuracy of different supervised machine learning algorithm which involves the importance of feature engineering term used in Machine learning.