Scholarlink

<p>Lead-free perovskite solar cells (PSCs) have gained significant interest in recent years due to their eco-friendly nature and cost-effectiveness. Among the promising lead-free absorber materials for perovskite solar cells (PSCs), the cesium tin chloride (CsSnCl<sub>3</sub>) has been recognized as one of the best alternatives. In this work, the photovoltaic (PV) characteristics of HTL-free CsSnCl<sub>3</sub>-based PSCs have been simulated using the solar cell capacitance simulator (SCAPS-1D), and the impact of different electron transport layers (ETLs) has been explored. The best device configuration was found to be Au/CsSnCl<sub>3</sub>/TiO<sub>2</sub>/TCO/Front Contact, achieving a power conversion efficiency of 23.21%. The open-circuit voltage (<i>V</i><sub>OC</sub>) was found to be 1.07&#xa0;V, while the short-circuit current density (<i>J</i><sub>SC</sub>) and fill factor (FF) were found to be 26.20&#xa0;mA/cm<sup>2</sup> and 88.10%, respectively. To predict the performance of the PSCs rapidly and effectively, a Histogram-Based Gradient Boosting Regressor (HistGBR) model has been created using seven significant parameters. The model was tested using a set of 11,520 data points and was assessed using the fivefold cross-validation method. The accuracy of the model was found to be satisfactory, achieving a coefficient of determination (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\({R}^{2}\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mrow> <mi>R</mi> </mrow> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation>) score of 0.999 for both the training and validation sets. In addition to that, SHapley Additive exPlanations (SHAP) values showed that the three most important parameters were electron affinity (<i>χ</i>), acceptor density (<i>N</i><sub>A</sub>), and bulk defect density (<i>D</i><sub>bulk</sub>), which together contributed to more than 76.05% of the model’s accuracy. The machine learning (ML) model was executed in both local using the Jupyter environment and on cloud, using Amazon Web Services (AWS) SageMaker. The results obtained in this research work demonstrate the potential of using interpretable and data-driven approaches for performance prediction and feature analysis in HTL-free and low-cost PSC technologies.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Cloud-based machine learning approach for performance prediction of HTL-free CsSnCl3 perovskite solar cells

  • 期刊论文

  • 发表日期：

    2026年4月18日

查看全文
  • Aditi Thakur,
  • Dhawan Singh

摘要

Lead-free perovskite solar cells (PSCs) have gained significant interest in recent years due to their eco-friendly nature and cost-effectiveness. Among the promising lead-free absorber materials for perovskite solar cells (PSCs), the cesium tin chloride (CsSnCl3) has been recognized as one of the best alternatives. In this work, the photovoltaic (PV) characteristics of HTL-free CsSnCl3-based PSCs have been simulated using the solar cell capacitance simulator (SCAPS-1D), and the impact of different electron transport layers (ETLs) has been explored. The best device configuration was found to be Au/CsSnCl3/TiO2/TCO/Front Contact, achieving a power conversion efficiency of 23.21%. The open-circuit voltage (VOC) was found to be 1.07 V, while the short-circuit current density (JSC) and fill factor (FF) were found to be 26.20 mA/cm2 and 88.10%, respectively. To predict the performance of the PSCs rapidly and effectively, a Histogram-Based Gradient Boosting Regressor (HistGBR) model has been created using seven significant parameters. The model was tested using a set of 11,520 data points and was assessed using the fivefold cross-validation method. The accuracy of the model was found to be satisfactory, achieving a coefficient of determination ( \({R}^{2}\) R 2 ) score of 0.999 for both the training and validation sets. In addition to that, SHapley Additive exPlanations (SHAP) values showed that the three most important parameters were electron affinity (χ), acceptor density (NA), and bulk defect density (Dbulk), which together contributed to more than 76.05% of the model’s accuracy. The machine learning (ML) model was executed in both local using the Jupyter environment and on cloud, using Amazon Web Services (AWS) SageMaker. The results obtained in this research work demonstrate the potential of using interpretable and data-driven approaches for performance prediction and feature analysis in HTL-free and low-cost PSC technologies.