Adaptive Surrogate Model Assisted Swarm Intelligence for Parameter Inversion of complex hydrological models

Abstract

Parameter inversion in hydrological models aims to estimate parameters from observed data, improving accuracy and understanding of the system. This process typically involves optimization algorithms to identify optimal parameter combinations, often resulting in significant computational costs due to the necessity for numerous model runs, particularly in complex hydrological models. To address this challenge, this study introduces the Adaptive Surrogate Model Assisted Swarm Intelligence (ASMA-SI) framework. ASMA-SI uses the iterative traces of swarm intelligence (SI) as a training sample set, fostering a tightly coupling between SI and the surrogate model while minimizing computational demands and enhancing search efficiency. The framework was applied to enhance three prominent SI algorithms: Particle Swarm Optimization (PSO), Grey Wolf Optimizer (GWO), and Whale Optimization Algorithm (WOA). Synthetic experiments and a case study were conducted to evaluate the inversion efficacy of ASMA-SI. In the synthetic experiments, ASMA-SI demonstrated faster convergence to the ‘true value’, while in the real-world case study, it outperformed in nearly all of the nine test groups, achieving better average performance metrics.