Data Generation Approaches to Detect Abnormal Conditions in Water Distribution Systems

Abstract

The water distribution system (WDS) abnormal condition detection is a technology that identifies WDS failure (e.g., pipe bursts) based on abnormal hydraulic behaviors. Its performance in detecting abnormal conditions is impacted by the quantity and quality of the collected hydraulic data. The hydraulic data for normal operation conditions could be provided using the measurements collected by smart meters and various other sensors. However, the abnormal condition hydraulic data cannot be obtained except by means of field tests and irregular system failures. Therefore, this study proposes three data generation approaches that utilize the following: (1) a hydraulic simulation model; (2) real-field WDS abnormal condition data (e.g., field tests, past failure data); and (3) both (1) and (2), simultaneously. Subsequently, the J-town network was applied in order to verify the proposed approaches. It was found that these approaches can generate synthetic failure data when considering the characteristics of actual events of failure and various other scenarios.