Sensitivity analysis to improve water distribution system optimisation

Abstract

This study provides a sensitivity analysis of water distribution systems using Sobol's method, with the intent of improving the optimisation procedure for complex design problems. Generally, sensitivity analysis can enhance optimal design by reducing the size of search spaces, guiding the optimisation process, identifying the most appropriate system performance metrics, and determining the key uncertainty sources in the case of stochastic design. Sobol's method is a global sensitivity analysis method that provides detailed information on main parameter effects and their interactions. Two Sobol sensitivity indices (first-order and total-order indices) are computed using the Latin Hypercube sampling method. Sensitivities have been quantified for a number of different performance metrics, including total pressure deficit, maximum pressure deficit, minimum surplus pressure, resilience, probabilistic robustness, and a specific risk measure. Two case studies are investigated: the New York Tunnels network and the Anytown network. Results show that different metrics have a distinctive response to different decision variables, highlighting the need for careful selection of the metrics in the design process. Significant interactions exist between different variables when analyzing risk and resilience in the case of New York Tunnels network, and for the minimum surplus head in the case of Anytown network. In addition, a large portion of decision variables have little impact on many of the metrics, offering a possible reduction of search space size when solving the optimal design problem. © 2012 ASCE.