{"title":"Machine learning for advanced characterisation of silicon photovoltaics: A comprehensive review of techniques and applications","authors":"Yoann Buratti , Gaia M.N. Javier , Zubair Abdullah-Vetter , Priya Dwivedi, Ziv Hameiri","doi":"10.1016/j.rser.2024.114617","DOIUrl":null,"url":null,"abstract":"<div><p>Accurate and efficient characterisation techniques are essential to ensure the optimal performance and reliability of photovoltaic devices, especially given the large number of silicon solar cells produced each day. To unlock valuable insights from the amount of data generated during the characterisation process, researchers have increasingly turned to different machine learning (ML) techniques. In this review, advances in ML applications for silicon photovoltaic (PV) characterisation from 2018 to 2023, including device investigation, process optimisation, and manufacturing line assessment are examined. Additionally, studies on deep learning techniques for luminescence-based measurements, such as defect classification, detection, and segmentation, which can help manufacturers identify potential reliability issues are explored. Despite the abundance of ML applications, it is emphasised that the lack of both publicly available datasets and the uniform use of ML metrics poses a significant challenge for researchers to benchmark their frameworks and achieve consistent and accurate results. In advancing ML applications in PV, future research should focus on improving model interpretability, balancing speed and accuracy, understanding computational demands, and integrating niche applications into a unified framework. Lastly, industry involvement and interdisciplinary collaboration among experts in solar energy, data science, and engineering are vital in tailoring ML solutions and enhancing innovation in addressing various challenges in the PV field.</p></div>","PeriodicalId":418,"journal":{"name":"Renewable and Sustainable Energy Reviews","volume":null,"pages":null},"PeriodicalIF":16.3000,"publicationDate":"2024-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.sciencedirect.com/science/article/pii/S1364032124003435/pdfft?md5=ae0394c4691b59afc69fbc07b10f9b06&pid=1-s2.0-S1364032124003435-main.pdf","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Renewable and Sustainable Energy Reviews","FirstCategoryId":"1","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S1364032124003435","RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"ENERGY & FUELS","Score":null,"Total":0}
引用次数: 0
Abstract
Accurate and efficient characterisation techniques are essential to ensure the optimal performance and reliability of photovoltaic devices, especially given the large number of silicon solar cells produced each day. To unlock valuable insights from the amount of data generated during the characterisation process, researchers have increasingly turned to different machine learning (ML) techniques. In this review, advances in ML applications for silicon photovoltaic (PV) characterisation from 2018 to 2023, including device investigation, process optimisation, and manufacturing line assessment are examined. Additionally, studies on deep learning techniques for luminescence-based measurements, such as defect classification, detection, and segmentation, which can help manufacturers identify potential reliability issues are explored. Despite the abundance of ML applications, it is emphasised that the lack of both publicly available datasets and the uniform use of ML metrics poses a significant challenge for researchers to benchmark their frameworks and achieve consistent and accurate results. In advancing ML applications in PV, future research should focus on improving model interpretability, balancing speed and accuracy, understanding computational demands, and integrating niche applications into a unified framework. Lastly, industry involvement and interdisciplinary collaboration among experts in solar energy, data science, and engineering are vital in tailoring ML solutions and enhancing innovation in addressing various challenges in the PV field.
准确高效的表征技术对于确保光伏设备的最佳性能和可靠性至关重要,尤其是考虑到每天生产的硅太阳能电池数量巨大。为了从表征过程中产生的大量数据中获得有价值的见解,研究人员越来越多地转向不同的机器学习(ML)技术。在这篇综述中,研究了 2018 年至 2023 年硅光伏(PV)表征的 ML 应用进展,包括设备调查、工艺优化和生产线评估。此外,还探讨了基于发光测量的深度学习技术研究,如缺陷分类、检测和分割,这些技术可以帮助制造商识别潜在的可靠性问题。尽管有大量的 ML 应用,但需要强调的是,缺乏公开可用的数据集和统一使用的 ML 指标,对研究人员基准测试其框架并获得一致、准确的结果构成了重大挑战。在推进光伏领域的 ML 应用方面,未来的研究应侧重于提高模型的可解释性、平衡速度和准确性、了解计算需求以及将利基应用集成到统一框架中。最后,太阳能、数据科学和工程领域专家的行业参与和跨学科合作对于定制 ML 解决方案和加强创新以应对光伏领域的各种挑战至关重要。
期刊介绍:
The mission of Renewable and Sustainable Energy Reviews is to disseminate the most compelling and pertinent critical insights in renewable and sustainable energy, fostering collaboration among the research community, private sector, and policy and decision makers. The journal aims to exchange challenges, solutions, innovative concepts, and technologies, contributing to sustainable development, the transition to a low-carbon future, and the attainment of emissions targets outlined by the United Nations Framework Convention on Climate Change.
Renewable and Sustainable Energy Reviews publishes a diverse range of content, including review papers, original research, case studies, and analyses of new technologies, all featuring a substantial review component such as critique, comparison, or analysis. Introducing a distinctive paper type, Expert Insights, the journal presents commissioned mini-reviews authored by field leaders, addressing topics of significant interest. Case studies undergo consideration only if they showcase the work's applicability to other regions or contribute valuable insights to the broader field of renewable and sustainable energy. Notably, a bibliographic or literature review lacking critical analysis is deemed unsuitable for publication.