Residual Saltwater Desalination Behind Subsurface Dams in an Unconfined Aquifer Using Hierarchically Gated Recurrent Neural Network

ABSTRACT

This study proposes an intelligent method to desalinate residual saltwater trapped behind subsurface dams in unconfined aquifers, addressing slow natural desalination and overlooked tidal effects. The proposed technique combines Hierarchically Gated Recurrent Neural Network (HGRNN) and Kookaburra Optimization Algorithm (KOA), termed as the HGRNN‐KOA technique. Input data are collected from the Iran Water Resources Management Company. Adaptive Robust Cubature Kalman Filtering (ARCKF) is applied to preprocess the data. The HGRNN predicts saltwater wedge migration, with KOA optimizing its weights. The proposed technique was evaluated and compared with existing techniques using the MATLAB platform. It demonstrated superior performance compared to approaches like Jellyfish Search (JS) algorithm, K‐nearest neighbour (KNN) and Bayesian optimization (BO) algorithm. The proposed approach achieved 98% efficiency in accurately estimating residual saltwater wedge length (RSWL) and residual total salt mass (RTSM), outperforming the existing approaches. The results demonstrate the approach's accuracy and reliability for managing saltwater intrusion and guiding subsurface dam design.