A data-driven method to construct prediction model of solar stills

The interdisciplinary field between solar desalination and machine learning is the subject of a cutting-edge study. Generally, the studies treat data acquisition and model construction as independent processes, leading to problems such as insufficient dataset size or resource wastage. This study proposes a data-driven method that integrates data acquisition with model construction processes. By using the Bayesian optimization algorithm, the method accelerates the convergence of model accuracy. By comparing the results of 100 pairs of simulations, it is found that the models using the data-driven method are more accurate than traditional expert-driven methods in 70% of compared results. Additionally, when it makes a model with the mean absolute percentage error as 5%, the proposed data-driven method requires 220 additional data on average, compared to 258 with the traditional expert-driven method, representing a 14.7% reduction. This work offers new ways and a broad application of the interdiscipline between solar desalination and machine learning.