Quantum and classical machine learning for perovskite bandgap learning and solar cell performance analysis

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

Abstract Perovskite materials have shown great potential for photovoltaic applications because of their excellent optoelectronic properties and tunable chemical composition. Classical machine learning (CML) has become an important method in this field, while quantum machine learning (QML) has shown preliminary value in perovskite classification tasks. Based on the Perovskite Database, this study constructs a high-quality experimental dataset. In the bandgap study, the composition-based feature vector (CBFV) method is introduced, and three key descriptors are identified through feature importance analysis and ablation experiments, achieving high prediction accuracy. These descriptors are then applied to QML classification and show good potential in bandgap interval classification. In the study of power conversion efficiency (PCE), material, device structure, and process information are integrated, and multi-task learning is used to improve prediction performance. In the PCE interval classification task, hybrid encoding and long-tail filtering reduce high-dimensional feature redundancy. Limited by qubit number and local simulation resources, this work adopts feature selection, grouped encoding, and data re-uploading strategies, and the results outperform CML methods. Multiple interpretability methods are further used to reveal the relationships between key factors and performance.

Links and sources

Comments