Performance prediction of solar still with a high-frequency ultrasound waves atomizer using random vector functional link/heap-based optimizer

Abstract

In this work, the thermal performance of a single slop solar still (SS) using a high-frequency ultrasound waves atomizer (HFUWA) as a humidifier was predicted through developing an artificial intelligence model. To minimize or avoid mathematical analysis or performing costly experimental work, a set of machine-learning (ML) algorithms were performed to predict the SS productivity and basin water temperature. A powerful ML algorithm called random vector functional link network (RVFL) is optimized using three different advanced metaheuristic optimizers, namely sine cosine algorithm (SCA), manta ray foraging optimizer (MRFO), heap-based optimizer (HBO). These optimizers are used to obtain the optimal parameters of RVFL model that maximize its prediction accuracy. By using experimental data, the algorithms; RVFL, R-SCA, R-MRFO and R-HBO, were trained and tested. The operational variables such as number of atomizers, water depth, and on-off time were used as input variables of the algorithms. The appropriate meteorological variables: atmospheric temperature, speed of wind, and solar intensity were considered as the input parameters. Results indicated that the R-HBO has a high ability to find out the process responses as a function of inputs in nonlinear relationship. RVFL-HBO achieves low RMSE, MRE, MAE and COV of 44.840, 0.512, 35.497 and 48.539 for water yield and 6.660, -0.019, 5.090 and 13.856 for water temperature. The low values of RMSE, MAE, MRE, and COV obtained by RVFL-HBO indicate its high accuracy over other models. Hence, it can be considered as the best choice for modeling the water desalination process in SS.