An optimization framework for large water distribution systems based on complex network analysis

Abstract

The major task of water distribution networks (WDNs) is to reliably supply water in sufficient quantity and quality. Due to the complexity in design and operation of WDNs, and to ensure a reliable level of service with minimum costs, multi-objective design approaches are used which are usually rely on evolutionary algorithms. However, for large WDNs the decision variable space increases exponentially. When considering multiple objectives (e.g., resilience, costs, water quality), for complex, large (real) WDNs with several thousand decision variables, evolutionary algorithms are practically infeasible to apply. With complex network analysis mathematical graphs of WDNs can be analysed very computationally efficient and therefore such an approach is especially suitable for analysing large spatial transport networks. Recently, based on complex network, a highly efficient approach for Pareto-optimal design of WDNs was developed. Based on topological features and a customized graph measure for the demand distribution (demand edge betweenness centrality), a graph-based multi-objective design approach was developed, which outperformed the results of an evolutionary algorithm regarding the quality of solutions and computation time (factor 105 faster). Further, also based on complex network analysis, a highly efficient surrogate method for assessing water quality in large WDNs was developed (2.4∙105 times faster than extended period simulation Epanet2). In this paper, these two approaches based on complex network analysis: (1) two objective optimization model and (2) the graph-based water quality model, are combined in a novel graph optimization framework which is especially suitable for complex, large (real) WDNs. The applicability of this very computationally efficient, novel approach is shown on a real case studies with 4,000 decision variables for which the results are be obtained within 18.5 seconds of computation time, while with a state-of-the-art evolutionary algorithm it took more than 8 weeks.