Abstract
There has been a growing trend for developing predictive solar-desalination models. However, forecasting productivities of solar stills of different designs remains a challenge. Herein, we developed predictive machine learning (ML) models for predictions of a double-slope still productivity based on experimental results. Trained datasets were taken from earlier designed passive and/or active solar stills used to treat brackish/wastewater with 45% TDS. FGSVM, EBoT, and SEGPR regression models showed the least possible MSE's (<138) indicating their reliability to accurately predict distillate amounts in double-slope still designs. The highest accuracy of SEGPR trained model with (R^{2}=1) and very low RMSE <8 shows its promise in predicting the performance of such similar solar-desalination systems. The novelty of this work is associated with paving the way towards creating a unified theoretical model that would provide the key to maximize still efficiencies and distillate-water outputs from supervised ML models allowing tuning of the correct parameters.