跳到正文
Cell Reports Physical Science· Muppana·收录于· 6 天前

A critical audit and methodological roadmap for machine learning in perovskite solar cells

A critical audit and methodological roadmap for machine learning in perovskite solar cells

原文摘要

出版方登记 · OpenAlex

Perovskite solar cells (PSCs) have advanced from 3.8% to 27.01% power conversion efficiency (PCE) in 17 years, while their lead-free analogs have reached 17.1%, and perovskite/silicon tandems exceeded 34%. Machine learning (ML) is credited with accelerating much of this progress, yet its evaluation is rarely examined. Here, we report a critical audit of 80 ML-PSC studies from 2024 to 2026. Among these, 89% report no error analysis on any sub-range of PCE, none report nested cross-validation, 48% rely on a single train/test split, and only 9% predict stability over time. Models that appear accurate overall can therefore fail in the sparse high-PCE region, which is the region most relevant to commercialization. We reframe high-PCE prediction as an imbalanced regression problem, establish imbalance-aware techniques, recommend an evaluation workflow, and propose a five-pillar roadmap. Further progress depends as much on shared evaluation and reporting standards as on new algorithms.

页面未能读取全文,以上为出版方登记的摘要;方法与图表见原文

来源:Cell Reports Physical Science · doi.org