Determining the Accuracy of Automated Calibration of Pipe Network Models

Abstract

Methods to automatically calibrate water distribution system models have been available for some time but it is very difficult to prove that any method is correct. Since at any one time the ability to know all the usage and flow conditions in a real system is impossible, obtaining all of the data needed in a real water distribution system to obtain an accurate and complete data set for model calibration is unrealistic. To test the ability of automated calibration methods to predict the actual conditions in a water system a laboratory scale physical model of a water distribution system was constructed and an automated water distribution model calibration program, employing genetic algorithms, was used to calibrate the model of that system. The results indicated that the automated calibration methods worked well in estimating pipe roughness, demands and locating closed valves. More specifically, the automated calibration model exactly matched the measured flows and pressures in the system. It was able to identify whether a valve was closed and where the demands were located. If given sufficient data, it was able to identify pipe roughness. The only problems occurred when the number of unknowns greatly exceeded the number of measurements. The model worked equally well regardless of whether the head loss equation used was the Hazen-Williams, Darcy-Weisbach or Manning equation. In all, automated calibration was successful. The paper describes how the lab data were collected, and how the calibration program matched the lab data and provides some suggestions for users of an automated water distribution calibration model.