<p>This study presents an integrated SCAPS-1D simulation and explainable machine learning framework for the performance optimization of lead-free <InlineEquation ID="IEq3"><EquationSource Format="TEX">\(\hbox {Cs}_{2}\hbox {AgInBr}_{6}\)</EquationSource></InlineEquation> double perovskite solar cells (PSCs). A comprehensive dataset was generated through SCAPS-1D simulations by systematically varying key device parameters, including absorber thickness, bandgap, doping concentration, and defect density. The generated dataset was subsequently employed to train and evaluate multiple machine learning models, namely Linear Regression (LIR), Quadratic Regression (QR), Support Vector Regression (SVR), Random Forest Regression (RFR), eXtreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), Multi-Layer Perceptron (MLP), and Gaussian Process Regression (GPR). Among the investigated models, GPR demonstrated the highest predictive capability, achieving an <InlineEquation ID="IEq4"><EquationSource Format="TEX">\(R^2\)</EquationSource></InlineEquation> score of 0.9999, a root mean square error (RMSE) of 0.0310, and a cross-validation score of 0.9998. To enhance model interpretability, SHapley Additive exPlanations (SHAP) analysis was employed to quantify the relative importance and directional influence of critical design variables on photovoltaic performance metrics. The SHAP-guided analysis identified defect density, absorber thickness, bandgap and doping concentration as the dominant factors governing device behavior and facilitated the determination of optimal parameter ranges. Under optimized simulation conditions, the proposed ITO/<InlineEquation ID="IEq5"><EquationSource Format="TEX">\(\hbox {Ag}_{2}\hbox {S}\)</EquationSource></InlineEquation>/<InlineEquation ID="IEq6"><EquationSource Format="TEX">\(\hbox {Cs}_{2}\hbox {AgInBr}_{6}\)</EquationSource></InlineEquation>/<InlineEquation ID="IEq7"><EquationSource Format="TEX">\(\hbox {Cu}_{2}\hbox {MnSnS}_{4}\)</EquationSource></InlineEquation>/Pt device architecture achieved a short-circuit current density (<InlineEquation ID="IEq8"><EquationSource Format="TEX">\(J_{sc}\)</EquationSource></InlineEquation>) of 31.696 mA/cm<InlineEquation ID="IEq9"><EquationSource Format="TEX">\(^{2}\)</EquationSource></InlineEquation>, an open-circuit voltage (<InlineEquation ID="IEq10"><EquationSource Format="TEX">\(V_{oc}\)</EquationSource></InlineEquation>) of 1.196 V, a fill factor (FF) of 88.14%, and a power conversion efficiency (PCE) of 33.41%. The corresponding GPR predictions showed excellent agreement with the SCAPS-1D results, demonstrating the reliability of the developed predictive framework. While the reported performance represents an idealized simulation scenario, the proposed approach provides valuable insights into parameter optimization and offers an efficient pathway for accelerating the design and development of lead-free double perovskite photovoltaic devices.</p>

SCAPS-1dD and explainable machine learning framework for performance optimization of lead-free \(\hbox {Cs}_{2}\hbox {AgInBr}_{6}\) double perovskite solar cells

期刊论文
出版模式：

开放获取
发表日期：

2026年8月1日

查看全文

Rupashree Dutta,
Deepika Mishra,
Dayanand Ray,
Akash Sharma

摘要

This study presents an integrated SCAPS-1D simulation and explainable machine learning framework for the performance optimization of lead-free \(\hbox {Cs}_{2}\hbox {AgInBr}_{6}\) double perovskite solar cells (PSCs). A comprehensive dataset was generated through SCAPS-1D simulations by systematically varying key device parameters, including absorber thickness, bandgap, doping concentration, and defect density. The generated dataset was subsequently employed to train and evaluate multiple machine learning models, namely Linear Regression (LIR), Quadratic Regression (QR), Support Vector Regression (SVR), Random Forest Regression (RFR), eXtreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), Multi-Layer Perceptron (MLP), and Gaussian Process Regression (GPR). Among the investigated models, GPR demonstrated the highest predictive capability, achieving an \(R^2\) score of 0.9999, a root mean square error (RMSE) of 0.0310, and a cross-validation score of 0.9998. To enhance model interpretability, SHapley Additive exPlanations (SHAP) analysis was employed to quantify the relative importance and directional influence of critical design variables on photovoltaic performance metrics. The SHAP-guided analysis identified defect density, absorber thickness, bandgap and doping concentration as the dominant factors governing device behavior and facilitated the determination of optimal parameter ranges. Under optimized simulation conditions, the proposed ITO/\(\hbox {Ag}_{2}\hbox {S}\)/\(\hbox {Cs}_{2}\hbox {AgInBr}_{6}\)/\(\hbox {Cu}_{2}\hbox {MnSnS}_{4}\)/Pt device architecture achieved a short-circuit current density (\(J_{sc}\)) of 31.696 mA/cm\(^{2}\), an open-circuit voltage (\(V_{oc}\)) of 1.196 V, a fill factor (FF) of 88.14%, and a power conversion efficiency (PCE) of 33.41%. The corresponding GPR predictions showed excellent agreement with the SCAPS-1D results, demonstrating the reliability of the developed predictive framework. While the reported performance represents an idealized simulation scenario, the proposed approach provides valuable insights into parameter optimization and offers an efficient pathway for accelerating the design and development of lead-free double perovskite photovoltaic devices.