Abstract
Tubular solar still is a simple light-weight desalination unit with a large condensing surface compared with other types of solar stills. Regrettably, it suffers from the low water yield like other types of solar stills. In this work, two main research themes are studied. The first is enhancing the water yield and thermal efficiency of tubular solar still by providing the absorber plate with an electrical heater powered by a solar photovoltaic panel. The performance of the modified solar still is evaluated based on its water yield as well as energy and exergy efficiencies. The second is developing a fine-tuned artificial intelligent model to predict the thermal efficiency and water yield of the solar still. The fine-tuned model consists of a traditional artificial neural network model optimized by a meta-heuristic optimizer called humpback whale optimizer. The prediction accuracy of the developed model is compared with that of the standalone artificial neural network model and an optimized model using a traditional particle swarm optimizer. The results showed that the conventional tubular solar still produces an average accumulated water yield of 2.58 L/m2/day, while the modified tubular solar still produces an average accumulated water yield of 3.41 L/m2/day with 31.85% improvement. The daytime energy efficiency of the modified tubular solar still is 38.61%, but for the conventional one is only 30.67%. Moreover, the new developed model has the highest prediction accuracy among other investigated models. The optimized model using humpback whale optimizer has the highest correlation coefficient ranges between 0.983 and 0.999, the optimized model using particle swarm optimizer has a moderate correlation coefficient between 0.969 and 0.987, and standalone model has the lowest ranges between 0.594 and 0.937. These results revealed the vital role of the electrical heater in enhancing the thermal performance of the solar still and the important role of humpback whale optimizer for improving the prediction accuracy of the traditional neural network models.