Scholarlink

<p>In this simualtion based investigation, we have built a triple absorber layer of FASnI<sub>3</sub>, CsSnCl<sub>3</sub>, Cs<sub>2</sub>AgBiBr<sub>6</sub> because of their high absorption coefficient, low recombination rate and high carrier mobility. Numerical calibration of the built device (FTO/TiO<sub>2</sub>/FASnI<sub>3</sub>/CsSnCl<sub>3</sub>/ Cs<sub>2</sub>AgBiBr<sub>6</sub>/CBTS) is performed using SCAPS-1D software. The varying bandgap profile of absorbers is strategically used to enhance solar spectral absorption by capturing photons across a wider solar spectrum, thereby reducing transmission losses, improving charge carrier utilization and suppeior charge extraction. This work demonstrates that increasing absorber thickness and doping density of absorbers up to an optimal point significantly enhances power conversion efficiency (PCE) by promoting higher photogenerated carrier density and allowing sufficient diffusion lengths for carrier to move, which in turn reduce recombination losses. Physical calibration has led to acheivment of overall PCE of 26.67%, fill factor (FF) of 85.90%, open circuit voltage (V<sub>OC</sub>) of 1.15&#xa0;V and J<sub>SC</sub> of 26.91&#xa0;mA/cm<sup>2</sup>. Different Machine Learning (ML) models in particular Random Forest (RF), Support Vector Regression (SVR), eXtreme Gradient Boosting (XGBoost) and Stacking (SVR + RF) models are implemented for prediction of PCE for simuntanously varying absorber thickness. The thickness of these three absorber layers (FASnI<sub>3</sub>/CsSnCl<sub>3</sub>/ Cs<sub>2</sub>AgBiBr<sub>6</sub>) are used as input parameter to train ML models for prediction of target variable that is PCE. The analysis of SHAP plot is conducted to identify which absorber layer thickness has most relevance in prediction process. In conculsion Stacking (SVR + RF) model has been found to be most accurate in predicting device PCE having mean squared error of 0.00103 approximately and coefficient determination (R<sup>2</sup>) of 0.99939.</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

Prediction of power conversion efficiency using machine learning in triple-absorber perovskite solar cells

  • 期刊论文

  • 发表日期：

    2025年7月29日

查看全文
  • Harsh Sudhakar,
  • Jaspinder Kaur,
  • Rikmantra Basu,
  • Ajay Kumar Sharma,
  • Jaya Madan,
  • Rahul Pandey

摘要

In this simualtion based investigation, we have built a triple absorber layer of FASnI3, CsSnCl3, Cs2AgBiBr6 because of their high absorption coefficient, low recombination rate and high carrier mobility. Numerical calibration of the built device (FTO/TiO2/FASnI3/CsSnCl3/ Cs2AgBiBr6/CBTS) is performed using SCAPS-1D software. The varying bandgap profile of absorbers is strategically used to enhance solar spectral absorption by capturing photons across a wider solar spectrum, thereby reducing transmission losses, improving charge carrier utilization and suppeior charge extraction. This work demonstrates that increasing absorber thickness and doping density of absorbers up to an optimal point significantly enhances power conversion efficiency (PCE) by promoting higher photogenerated carrier density and allowing sufficient diffusion lengths for carrier to move, which in turn reduce recombination losses. Physical calibration has led to acheivment of overall PCE of 26.67%, fill factor (FF) of 85.90%, open circuit voltage (VOC) of 1.15 V and JSC of 26.91 mA/cm2. Different Machine Learning (ML) models in particular Random Forest (RF), Support Vector Regression (SVR), eXtreme Gradient Boosting (XGBoost) and Stacking (SVR + RF) models are implemented for prediction of PCE for simuntanously varying absorber thickness. The thickness of these three absorber layers (FASnI3/CsSnCl3/ Cs2AgBiBr6) are used as input parameter to train ML models for prediction of target variable that is PCE. The analysis of SHAP plot is conducted to identify which absorber layer thickness has most relevance in prediction process. In conculsion Stacking (SVR + RF) model has been found to be most accurate in predicting device PCE having mean squared error of 0.00103 approximately and coefficient determination (R2) of 0.99939.