Neural network-assisted expensive optimisation algorithm for pollution source rapid positioning of drinking water

Abstract

Pollution source positioning is a complicated problem because urban water supply networks contain a huge number of nodes and it is also a computationally expensive problem. Surrogate model-based intelligent optimisation algorithms can effectively solve such problems. In this study, multiple offline neural network models were constructed using big data technology, which saves time otherwise needed for online model construction. Moreover, a variety of model management strategies are proposed and their validities are experimentally confirmed. Based on this, a neural network-assisted optimisation algorithm is proposed to rapid position of pollution source. The experimental results shown this novel algorithm can greatly reduce computing time while ensuring positioning accuracy.