Genetic algorithm hyper-parameter optimization using Taguchi design for groundwater pollution source identification

Abstract

Groundwater pollution has been a major concern for human beings, since it is inherently related to people's health and fitness and the ecological environment. To improve the identification of groundwater pollution, many optimization approaches have been developed. Among them, the genetic algorithm (GA) is widely used with its performance depending on the hyper-parameters. In this study, a simulation–optimization approach, i.e., a transport simulation model with a genetic optimization algorithm, was utilized to determine the pollutant source fluxes. We proposed a robust method for tuning the hyper-parameters based on Taguchi experimental design to optimize the performance of the GA. The effectiveness of the method was tested on an irregular geometry and heterogeneous porous media considering steady-state flow and transient transport conditions. Compared with traditional GA with default hyper-parameters, our proposed hyper-parameter tuning method is able to provide appropriate parameters for running the GA, and can more efficiently identify groundwater pollution.