Machine learning-driven development of semiconductor photothermal materials for interfacial solar evaporation

Solar water evaporation is regarded as a promising method to solve the global freshwater shortage due to its lower cost, green and sustainable. Semiconductors photothermal materials (SPMs), with modifiable bandgap, and high performance, exhibit excellent prospects for practical interfacial solar evaporation. However, the precise influence of SPMs properties on evaporation performance remains unclear, necessitating long time and large costs to develop by trial-and-error tests. Machine learning (ML) emerges as a data-driven approach to achieve feature classification and performance prediction in evaluating SPMs. In this work, four ML models were employed to analyze the database collected from the latest research results of SPMs. Particularly, the voting regression (VR) model displayed a predication accuracy R² of 0.9337 for predicting evaporation rate on the test set, while the R² of VR model is only moderate (0.724) for predicting evaporation efficiency on test set, which can be improved via dataset expansion in the future work. Shapley Additive Explanation analysis revealed the most impact feature for evaporation performance of SPMs. Moreover, experiments verified sodium borohydride (NaBH₄) doping titanium dioxide (TiO₂) to form oxygen vacancies under calcination temperature 700 °C, resulting in bandgap narrowing and enhanced light absorption. The TiO₂ sample at 700 °C exhibited best evaporation performance, with evaporation rates of 1.12, 1.62, 1.89 and 2.28 kg m⁻² h⁻¹ under 1sun, 2sun, 3sun and 4sun, respectively. This study highlights that ML strategies can offer an efficient guideline for the rational design and development of high-performance SPMs for solar-driven water evaporation.