Identifying Sampling Interval for Event Detection in Water Distribution Networks

Abstract

It is a generally adopted policy, albeit unofficially, to sample flow and pressure data at a 15-min interval for water distribution system hydraulic measurements. Further, for flow, this is usually averaged, whereas pressure is instantaneous. This paper sets out the findings of studies into the potential benefits of a higher sampling rate and averaging for flow and pressure measurements in water distribution systems. A data set comprising sampling at 5 s (in the case of pressure), 1 min, 5 min, and 15 min, both instantaneous and averaged, for a set of flow and pressure sensors deployed within two DMAs has been used. Engineered events conducted by opening fire hydrants/wash outs were used to form a controlled baseline detection comparison with known event start times. A data analysis system using support vector regression (SVR) was used to analyze the flow and pressure time series data from the deployed sensors and hence, detect these abnormal events. Results are analyzed over different sensors and events. The overall trend in the results is that a faster sampling rate leads to earlier event detection. However, it is concluded that a sampling interval of 1 or 5 min does not significantly improve detection to the point at which it is worth the added increase in power, communications, and data management requirements with current technologies. It was discovered that averaging pressure data can result in more rapid detection when compared with using the same instantaneous sampling rate. Averaging of pressure data is also likely to provide better regulatory compliance and provide improved data for EPS hydraulic modelling. This improvement can be achieved without any additional overheads on communications by a simple firmware alteration and hence, is potentially a very low cost upgrade with significant gains.