Inversion-based identification of DNAPLs-contaminated groundwater based on surrogate model of deep convolutional neural network

Abstract

Abstract This paper combines theoretical analysis with practical examples to examine outstanding issues in research on the inversion-based identification of dense non-aqueous phase liquids (DNAPLs) in groundwater. We first generalize the relevant geological and hydrogeological conditions to establish a conceptual model of groundwater contamination. We then use it to formulate a preliminary model of the contamination of groundwater by DNAPLs based on multi-phase flow to describe the mechanism of migration of these pollutants. Following this, a surrogate model is established by training and validating the deep convolutional neural network (DCNN) based on training samples and samples for verification. Finally, the surrogate model is embedded into an optimization model as an equality constraint and the particle swarm optimization (PSO) algorithm is used to solve it.