A multi-chain surrogate-assisted hybrid optimization framework for joint identification of groundwater contaminant sources and hydrogeological parameters

Abstract

Rapid and accurate identification of groundwater contaminant information and hydrogeological parameters is crucial for effective groundwater remediation and risk management. Within a simulation–optimization framework, this task is inherently posed as a mixed-variable optimization problem involving discrete parameters (e.g., source locations) and continuous ones (e.g., hydraulic heads, conductivities, and release fluxes). However, several challenges arise in this context. First, conventional optimization algorithms often exhibit slow convergence and unstable performance. Second, they typically require thousands of simulations to adequately explore the complex parameter space, resulting in prohibitive computational costs. To address these issues, this study develops a surrogate-assisted hybrid algorithm that integrates the Cooperative Search Algorithm (CSA) and Tabu Search (TS) within a synergistic multi-chain optimization framework, termed SA-CSA-TS. In each iteration, individual chains first perform independent CSA-based optimization to promote broad global exploration, after which they collaboratively refine source locations through a neighbourhood search guided by a shared tabu list. In addition, surrogate models equipped with a reconstruction strategy partially replace groundwater simulations, thereby substantially reducing the computational burden. Case studies reveal that the Radial Basis Function (RBF) outperforms other mainstream surrogate models in both accuracy and stability. Furthermore, comparative experiments confirm that the proposed SA-CSA-TS framework not only achieves higher solution accuracy but also significantly reduces computational demand, demonstrating strong potential for efficient groundwater contamination diagnosis.