Automated Calibration of Real Water Distribution Networks: City X Case Study

Abstract

The calibration of water distribution systems networks (WDN) is a complex non-linear problem that has traditionally been solved using manual trial-and-error methods. Over the past few decades, automated calibration techniques have been the subject of a great deal of research, and significant advancements in the field of WDN calibration have been achieved. Despite these advancements in calibration techniques, and the development and application of automated tools in numerous research settings, most practitioners still rely on engineering judgment and traditional calibration methods for WDN calibration. In an effort to bridge the gap between practitioners and researchers, the value of these advanced calibration approaches relative to traditional methods is validated and quantified in a real world scenario. City X’s WDN model has been manually calibrated by engineers familiar with the system. Their calibration procedure was mainly an expert-based approach using judgment and trial-and-error and did not rely on automated optimization tools. The purpose of this study is to resolve the corresponding calibration problem with optimization tools and compare the calibration solutions in terms of quality (closeness to measured data) and calibration parameter values. The calibration problem is posed as a multi-objective optimization problem and solved with the PA-DDS algorithm. The precise calibration objectives are roughly matched to the manual calibration objectives specified by the engineers who calibrated the model. Multi-objective optimization results are compared against the manually calibrated model, and the potential benefit of applying an automated approach is assessed. The utilization of automated calibration methods is shown to clearly assist in parameter error detection and model quality improvement. To further improve the quality of the calibrated model, the initial results from the automated calibration are used to guide a manual calibration effort. Upon the manual adjustment of macrocalibration parameters, the automated calibration tools are used again to quantify the parameter uncertainty. As such, the calibration approach is iterative, and based upon the effective application of automated calibration tools coupled with traditional trial-and-error methods. Several thousand unique solutions of approximately equal or greater quality than the existing manually calibrated model are generated. 89 of the most robust solutions are identified in an effort to parsimoniously capture the model parameter uncertainty. Ideally, this list of candidate solutions can be used in future operational studies to ensure that operational decisions are robust to a wide range of potential network conditions. Moreover, particularly uncertain model parameters are identified so as to focus future field studies, and data collection efforts.