Ensemble Learning Approaches for Solar Irradiance Prediction in Bangladesh

Forecasting solar irradiance plays a vital role in solar energy generation and can enhance the configuration and oversight of photovoltaic systems. Moreover, research on multi-step ahead forecasting remains limited, even though it is essential in improving the dispatching efficiency of photovoltaic systems. This study aims to forecast multi-step solar irradiance by devel-oping a prediction model incorporating multiple exogenous fea-tures. Hourly irradiance data from two locations in Bangladesh (i.e., Dhaka and Bogra) are utilized to evaluate the performance of several ensemble learning algorithms, each integrated with different feature selection techniques. Three technical metrics, including mean absolute error (MAE), root mean squared error (RMSE), and coefficient of determination$(\mathrm{R}^{2})$, were employed to evaluate the prediction accuracy of the ensemble algorithms. The Extra Trees (ET) Regressor outperforms others, achieving$\mathrm{R}^{2}$values of 0.859, 0.825, and 0.801 for 1, 2, and 3-hour forecasts in Dhaka, and 0.894, 0.850, and 0.832 in Bogra.