Predictive performance assessment of single and multi-junction solar cells: a machine learning approach

Engineering notes

Engineering notes will be added by the Power for Solar editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为光伏效率、钙钛矿太阳能电池、储能技术、太阳能热利用、BIPV、并网技术等高价值论文补充中文说明。

Original abstract

The photovoltaic landscape has diversified rapidly, yet a unified framework for quantitatively comparing and predicting device performance across material systems remains elusive; consequently, this study presents a machine learning-driven assessment of state-of-the-art solar cells, where performance metrics—including open-circuit voltage [Voc], short-circuit current density [Jsc], fill factor [FF], and active area—were extracted from recently compiled efficiency tables encompassing single-junction, multi-junction, and tandem architectures that span silicon, III-V compounds, perovskites, and chalcogenides, after which six regression algorithms were trained and evaluated for their capacity to predict each performance parameter, revealing that gradient boosting methods achieved a perfect coefficient of determination for Voc [R² = 1.00] and strong predictive power for Jsc [R² = 0.81], whereas a multilayer perceptron [MLP] neural network proved most effective for overall cell efficiency [R² = 0.86]; moreover, a feature importance analysis derived from the gradient boosting model showed that fill factor exerts the greatest influence on cell efficiency, followed by open-circuit voltage, thereby establishing a quantitative framework for anticipating device performance and providing data-driven directives for targeted optimization in photovoltaic research and development.

Links and sources

Comments