Optimizing Hyperparameters for Advanced Deep Neural Networks to Predict Solar Still Efficiency

Abstract

As the demand for sustainable water treatment solutions grows, passive solar distillation emerges as a promising technology for the efficient desalination of brackish water. This study explores the optimization of solar distillation systems for use in various sectors such as residential, agricultural, and industrial. The performance of these systems is influenced by several factors, including solar radiation, ambient conditions, wind speed, and the design of the system itself. By adopting cutting-edge machine learning techniques, this research presents a novel method employing Deep Neural Networks, with a focus on the Multilayer Perceptron (MLP) model, to enhance the accuracy of yield predictions. Through a detailed comparison of different hyperparameter optimization methods, the integration of Particle Swarm Optimization (PSO) with the MLP model was identified as the most effective approach. This combined PSO-MLP model, particularly when applied to a specific design of solar collectors, achieved remarkable results, highlighted by a Coefficient of Determination (COD) of 0.98167 and a Mean Squared Error (MSE) of 0.00006. The study illustrates the profound potential of advanced computational techniques in improving the efficiency of solar distillation systems, contributing valuable insights to the field of sustainable water purification.