Improving prediction of anti-reflective film-coated photovoltaic solar panel efficiency by integrating bayesian with machine learning algorithms

This manuscript explores machine learning models for predicting the efficiency of anti-reflective film-coated photovoltaic solar panels in the South Indian climate. Three models—standard artificial neural network (ANN), random forest algorithm, and multilinear regression—were developed and compared. 80% of the dataset was used for training and 20% for testing. The random forest model demonstrated superior effectiveness with a lower prediction error. Bayesian optimization refined both ANN and random forest models. Experiments yielded an average solar panel efficiency of 16.79%, with performance indicators (coefficient of determination) of 0.96966, 0.93466, 0.98419 and mean absolute percentage errors of 7.518, 10.658, 5.089%. Bayesian optimization improved the traditional ANN model by up to 36.2%. Random forest exhibited lower sensitivity to hyperparameters compared to ANN. Two important parameters such as coating thickness and solar insolation were identified by feature sensitivity analysis. In terms of accuracy and robustness, random forest outperformed in predicting anti-reflective film-coated photovoltaic solar panel efficiency.