Evaluating economic feasibility and machine learning-driven prediction for solar still desalination in Iran

Abstract

• Applied machine learning model to predict freshwater production. • Model predicted with high accuracy: R² 0.999, MAE 0.415, RMSE 0.572. • Freshwater production compared in Tehran and Shiraz. • Model predicted daily freshwater: Tehran 6688 ml, Shiraz 7847 ml. • The daily benefit ratio increases from 0.298 $ in Tehran to 0.349 $ in Shiraz. Water scarcity is a critical issue worldwide, particularly in arid regions where access to freshwater is limited. Solar stills provide a sustainable and cost-effective solution for desalinating saltwater to produce freshwater using renewable solar energy. Given the increasing global demand for freshwater, especially in countries like Iran, optimizing solar still performance is crucial. Tehran and Shiraz were selected for this study due to their high water stress, driven by large populations, and their significant potential for solar energy use due to high solar irradiation. This study applies various machine learning models to intelligently predict freshwater production from solar stills in these cities. Models such as Linear Regression (LR), Decision Tree (DT), Random Forest (RF), Bayesian Regression, and K-Nearest Neighbors (KNN) were evaluated. LR emerged as the most accurate model, with an R² of 0.999, a mean absolute error (MAE) of 0.415, and a root mean square error (RMSE) of 0.572. The model's predictions revealed substantial differences in potential freshwater production between Shiraz (6.76 L/m 2 day) and Tehran (5.77 L/m 2 day), reflecting regional variations in solar availability, also Shiraz demonstrates a reduction in the cost per litre from 0.029 $/L in Tehran to 0.024 $/L. These findings highlight the importance of location-specific solar desalination solutions and demonstrate the potential of integrating machine learning to optimize solar still systems for different geographic areas

Topics & Concepts