Performance prediction of multi-effect desalination (MED) plant: Experimental evaluation, machine learning modelling and optimization

Desalination is one of the most viable solutions for providing clean and sustainable drinking water in regions experiencing water scarcity. This research shows a thorough machine learning (ML) based approaches for optimization and performance prediction for a Multi effect desalination (MED) system. To evaluate the system’s performance, MED experiments were conducted at different temperatures which was set at 40 °C, 50 °C and 60 °C. A comparative study analysis of four ML models, Artificial Neural Network (ANN), Support Vector Regression (SVR), Random Forest (RF), and Gradient Boosted Regression (GBR) for the prediction and optimization of distillate production is carried out. Significant variations in predictive accuracy and robustness under various operating conditions were revealed by a comparison of the developed ML models. The relative strengths of the modelling approaches were highlighted by evaluation using R², MAE, and RMSE, which also showcases the importance of selecting the right model for accurate MED performance prediction. The best fit model was then further utilized for the input feature importance and optimization purposes. Feature importance analysis showed that hot water inlet temperature (T_{hot in}) was the most influential parameter followed by feed water temperature to the steam generator (T_feed in SG). These two parameters account for roughly 60.81% of the cumulative importance which indicated that optimizing and precisely controlling these variables should be given top priority to improve MED system performance. A maximum distillate production of 1.27 LPM was predicted by the ML-based optimization, which is a 223.52% increase over the average experimental distillate production of 0.39 LPM. This shows the great potential of data-driven optimization for significantly improving MED system performance.