فهرست نشریات علمی دسترسی آزاد ایران

This study presents an integrated approach that combines laboratory experiments, numerical modeling, and machine learning to enhance salinity prediction in coastal aquifers. A laboratory model simulated saltwater intrusion under controlled conditions, systematically varying water density (25, 35, 50 g/l), extraction rate (0.05, 0.12, 0.25 l/min), and extraction depth (12, 25, 36 cm) to examine hydrodynamic and geochemical interactions. Experimental data calibrated the SEAWAT numerical model, which generated additional scenarios for machine learning analysis. The study developed hybrid predictive models combining Support Vector Regression and Random Forest with Convolutional Neural Networks, optimized using the Golden Ball Optimization algorithm. Key input parameters, including relative water density, extraction rate, and well depth, were evaluated through a comprehensive statistical analysis. The evaluation results indicated that the RF-GBO-3 model, with a root mean square error (RMSE) of 0.039, exhibited the best performance among the models, while the SVR-GBO-3 model, with an RMSE of 0.056, also showed satisfactory performance. The Golden Ball Optimization algorithm enhanced model performance by effectively tuning critical parameters and capturing complex nonlinear relationships. These findings advance saltwater intrusion modeling by providing a robust framework that integrates physical experiments with data-driven techniques, offering improved tools for coastal water resource management.