Sensitivity-Oriented Clustering Method for Parameter Grouping in Water Network Model Calibration

As water network models contain a lot of parameters that are hard to directly obtain or be precisely measured, model calibration technologies are often used to estimate unknown parameters and ensure model accuracy. A main problem faced by calibration is the underdetermination due to the low ratio between available field measurements and parameters to be estimated. Parameter grouping is an effective approach to turn the calibration problem to even or over-determined, but current studies often group parameters manually according to engineering experience, which has low efficiency and accuracy, and even cause non-convergence of the calibration process. In this study, a Sensitivity-Oriented Clustering (SOC) method is proposed to automatically group unknown pipe roughness coefficients with promised convergency and promoted accuracy of the calibration process. SOC divides the grouping process into two stages. The first one focuses on high-sensitivity pipes to guarantee the convergence of calibration, and the second stage grouping attaches the remaining low-sensitivity pipes to the determined high-sensitivity groups to ensure estimation accuracy of these pipes. Two case studies are conducted to illustrate the application and effectiveness of the proposed grouping method. It has been proved that the SOC provides an intelligent way to group unknown parameters and can be combined with various calibration methods. The results show that SOC significantly improves the stability and accuracy of the calibration process, which makes it reliable and effective in practical application.