Machine learning-based model for predicting freshwater production and power consumption in solar-assisted desalination systems

Traditional modeling techniques frequently fail to address the complex, multi-variable optimization problem that arises when freshwater generation and energy efficiency are combined in greenhouse systems. Resolving this issue is essential to improving the sustainability of controlled agricultural settings, especially in areas with limited water resources or high energy consumption. In order to improve operational planning and system design, this study suggests a strong machine learning-based framework for precisely predicting power consumption and freshwater production in a greenhouse-integrated system. Five-fold cross-validation, hybrid Grey Wolf Optimizer (GWO) tuning, SHAP sensitivity analysis, and Taylor diagrams were used to assess a variety of machine learning models, such as XGBoost, CatBoost, SVR, MLP, KNN, and ElasticNet. The XGBoost-GWO model outperformed the others, obtaining the highest R² values (up to 0.9991) and the lowest RMSE (0.4933 for freshwater, 0.0311 for power). Plus, deep learning models such as LSTM and DNN show limited performance in freshwater prediction with high errors and longer runtimes, whereas XGBoost proves more accurate and computationally efficient for this application. Greenhouse width was found to be the most significant design parameter by feature importance and sensitivity analyses. Additionally, an ideal configuration that produced 99.80 m³ of freshwater per day with a mere 2.75 kWh/m³ energy consumption was found using a multi-objective optimization approach. This combined modeling and optimization method promotes resource efficiency and sustainable agriculture by providing a useful tool for designing greenhouse systems that have significant practical applications in resolving freshwater scarcity in arid and semi-arid areas.