Organic photovoltaic prediction model based on Bayesian optimization and explainable AI.

Area of Science:

• Renewable Energy
• Materials Science
• Computational Chemistry

Background:

• Photovoltaic technology is crucial for clean energy, but improving solar cell efficiency and cost-effectiveness faces challenges.
• Current methods often rely on empirical observations, limiting predictive power in complex energy chemistry.
• Machine learning offers a path to streamline predictions and discover novel materials for enhanced solar cell performance.

Purpose of the Study:

• To develop a novel hybrid-optimized multi-objective predictive model for key photovoltaic parameters.
• To accurately predict open-circuit voltage (Voc), current density (Jsc), fill factor (FF), and power conversion efficiency (PCE).
• To integrate Explainable Artificial Intelligence (XAI) for feature importance analysis.

Main Methods:

• A hybrid model combining Bayesian Optimization (BO) with ensemble Bootstrap Aggregating (Bagging) decision trees.
• Utilized Explainable Artificial Intelligence (XAI) via SHAP (Shapley Additive Explanations) for feature analysis.
• Evaluated model performance using correlation coefficient (r), coefficient of determination (R²), and Mean Square Error (MSE).

Main Results:

• Achieved high prediction accuracy with r=0.92, R²=0.82, and MSE=0.00172.
• Demonstrated efficient processing with short training (182.7 s) and inference times (0.00062 s).
• The BO-Bagging model showed a prediction speed of 2188.4 observations/sec and a model size of 10,740.4 KB.

Conclusions:

• The proposed BO-Bagging model accurately and efficiently predicts photovoltaic parameters.
• Feature importance analysis using XAI provides valuable insights into material properties affecting solar cell performance.
• This approach advances intelligent chemical applications in the renewable energy sector.