Interpretable machine learning insights of power conversion efficiency for hybrid perovskites solar cells

Abstract

Hybrid organic–inorganic perovskites (HOIPs) solar cells have presented broad application prospects in the photovoltaic field due to their high energy conversion efficiency, ease of preparation, and low production costs. With the flourishing development of artificial intelligence, machine learning (ML) has been recently used for novel HOIP designs. However, the practical application of ML models for the designing of HOIPs is hampered mainly due to the lack of interpretability. Herein, a data-driven interpretable ML approach is introduced to distill the universal simple descriptors for the power conversion efficiency (PCE) of HOIPs-based solar cells. It is highlighted that two descriptors consist of easily obtained parameters are proposed to accurately predict PCE, which are superior to the commonly used descriptor band gap (E_g). Remarkably, universal criterions for the high-throughput screening of HOIPs are proposed to accelerate the screening of HOIPs-based solar cells with high PCE performance. This work paves the way toward rapid and precise screening of efficient HOIPs-based solar cells using a data-driven interpretable ML approach.