Framework for Identification of Groundwater Contamination Source Based on Conditional Generative Adversarial Networks and Optimization Methods

Abstract

Accurate groundwater contamination source identification (GCSI) is critical for ensuring water resource security and management. However, solving the GCSI problem often faces challenges of insufficient identification accuracy and difficulty in quantifying uncertainty. To address these issues, we propose a framework based on a bidirectional mapping strategy, optimization methods, and an adaptive enhancement iterative process to simultaneously identify contamination source characteristics and model parameters. Specifically, the framework directly models the probability distribution of the variables to be identified based on a conditional generative adversarial network (CGAN), generates diverse samples, and thus achieves the quantification of uncertainty. Additionally, the innovative integration of CGAN and optimization methods enables the optimization algorithm to utilize its strong search capability to further refine the identification results. Application results on a hypothetical case show that the framework outperforms the use of optimization methods alone in terms of accuracy, efficiency, and reliability. The proposed framework enhances identification accuracy and explicitly quantifies uncertainty, providing a new solution for GCSI challenges.