摘要
The electrodialysis (ED) system is well-known for its efficient method for desalination, characterized by its long service life, high water recovery capacity, and low energy consumption. In the ED systems, an accurate approximation of the Separation Rate ( SR ) plays a critical role in optimizing freshwater production efficiency. Hence, this study proposes a novel Nested-Group Method of Data Handling (Nested-GMDH) model for predicting SR . Four variations of the nested Machine Learning (ML) models are developed utilizing different base learners, including (i) Shallow Neural Network (SNN), (ii) Deep Neural Network (DNN), (iii) Support Vector Regression (SVR), and (iv) Classification and Regression Tree (CART). Using a 5-fold cross-validation approach, the performance of these models was evaluated against four individual counterparts (SNN, DNN, SVR, and CART), standard GMDH, Homonuclear Molecules Optimization (HMO)-tuned GMDH, and three advanced ensemble methods (random forest, LSBoost, and XGBoost). The results, assessed via statistical metrics ( RMSE , R 2 , IA , & CE ) and a comprehensive TOPSIS ranking technique, clearly demonstrate the superior predictive capability of the proposed nested architecture. Specifically, the Nested-DNN model obtained the top rank with the highest accuracy across all criteria. Bland-Altman uncertainty analysis further validated the superior reliability of the developed nested ML methods and ensemble algorithms by specifically demonstrating the narrowest limits of agreement (±16.38%, spanning from −20.92% to + 11.83%) and minimal systematic bias (−6.25%) for the Nested-DNN. Future research may take advantage of the Nested-GMDH approach and apply it to multi-objective optimization problems such as minimizing specific energy consumption while maximizing SR from renewable energy sources.