Improving metamodel-based optimization of water distribution systems with local search

Abstract

Metamodels can be used to aid in improving the efficiency of computationally expensive optimization algorithms in a variety of applications, including water distribution system (WDS) design and operation. Genetic Algorithm (GA)-based optimization of WDSs is very computationally expensive to optimize a system in a practical amount of time for real-sized problems. A metamodel, of which Artificial Neural Networks (ANNs) are an example, is a model of a complex simulation model. It can be used in place of the simulation model where repeated use is necessary, such as when carrying out GA optimization. To complement the ANN-GA, six local search algorithms have been developed or applied in this research, with the aim of improving the performance of metamodel-based optimization of WDSs. All algorithms performed well, however, using computational intensity as a criterion with which to evaluate results, the best local search algorithms were Sequential Downward Mutation (SDM) and Maximum Savings Downward Mutation (MSDM).