Improving near real-time anomaly event detection and classification with trend change detection for smart water grid operation management

Abstract

ABSTRACT To improve the efficiency of Smart Water Grid (SWG) operation, it is of great importance to timely detect and classify anomaly events. Authors have developed a comprehensive solution framework for anomaly detection. However, the framework may detect undesirable false alarms, which will incur wasteful field investigation. Therefore, it is imperative to reduce the number of false alarms for real world SWG operation. In this paper, the anomaly event classification has been improved by an algorithm of trend change detection in two phases including (1) detecting trend change in single sensor, and (2) identifying the effective evaluation time window by correlating the flow anomaly trend period and pressure anomaly trend periods to filter the detected events. A real case study has been conducted to demonstrate the application of the proposed algorithm; the number of false alarms is reduced by as much as 31% without compromising the detection accuracy.