Simultaneous identification of groundwater contamination source and simulation model parameters based on the rime optimization algorithm.

Groundwater contamination source identification (GCSI) can accurately obtain characteristic information of contamination sources such as their locations and release histories, critical for contaminated site remediation and liability attribution. Among existing inversion methods, simulation optimization approaches face challenges including insufficient fitting accuracy of surrogate models and optimization algorithms being prone to fall into local optima. Based on the simulation optimization method, this study innovatively proposes a framework for solving GCSI problems. The proposed framework embeds a one-dimensional convolutional neural network (1DCNN) as an integrated surrogate component within the optimization model. The rime optimization algorithm (RIME) is employed to solve this composite optimization model, achieving simultaneous identification of both contamination source characteristics and aquifer parameters. The feasibility of the method is verified through a hypothetical case study. In a comparative analysis with fully connected neural network (FCNN) and support vector regression (SVR), the 1DCNN achieves an R value of 0.9998. Under ± 20% noise interference, the 1DCNN's R value remains above 0.9993. Compared with Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Energy Valley Optimizer (EVO), RIME demonstrates superior performance with an average relative error of 8.88% for single identification and maintains the average error of 5.88% across 100 repeated experimental trials. This stems from RIME's unique soft rime and hard rime search strategies, which enable effective escape from local minima and convergence to global optima. This method provides a new approach for solving GCSI problems and has potential for extension to problems such as multiphase contaminant transport and heterogeneous aquifer parameter identification.