A machine learning approach to assess the performance of solar still systems under Tunisian climate conditions
摘要
The limited productivity and efficiency of conventional solar stills remain a major challenge for sustainable desalination, particularly in arid regions where water demand is increasing despite abundant solar energy resources. Although several design modifications and modeling approaches have been proposed, the combined impact of absorber surface enhancement and advanced predictive modeling on solar still performance is still insufficiently explored. In this context, the present study evaluates the performance of conventional and modified pyramid solar stills through a combined experimental and computational approach. Experiments were conducted in Gabes, Tunisia, under real climatic conditions. An Artificial Neural Network (ANN) model, based on the Levenberg–Marquardt algorithm, was developed to predict freshwater productivity, energy efficiency, and exergy efficiency using key input variables, including absorber plate area, solar radiation, water temperature, vapor temperature, and wind speed. The model was trained, validated, and tested using experimental data to ensure high accuracy and reliability. The results demonstrate that increasing the absorber surface area significantly enhances thermal behavior, leading to higher temperature levels and improved system performance. The modified configuration achieved a freshwater productivity of 3440 ml/m2 compared to 3064 ml/m2 for the conventional system. In addition, the integration of 3D response surface analysis provided deeper insight into the interactions between operating parameters, enabling better optimization of system design and performance. Overall, this study highlights the effectiveness of combining experimental investigation with ANN-based predictive modeling to improve solar still efficiency and provides practical guidance for the development of high-performance solar desalination systems in arid environments.