Predictions of solar power using ensemble machine learning techniques

Abstract

Predicting solar power production accurately is becoming more and more crucial for efficient power management and the grid's integration of renewable energy sources. Using data from an Australian photovoltaic (PV) power station, this study employs a variety of machine learning (ML) ensemble techniques, such as gradient boosting (GB), random forest (RF), and extreme gradient boosting (XGBoost), to forecast solar power production. ML models are developed utilizing pertinent information from electricity and meteorological data in order to forecast solar power. The predictive performance of trained ML models is verified in terms of metrics like mean absolute error (MAE), root mean square error (RMSE), and correlation coefficient (R2). With higher R2 values and lower error results (MAE and RMSE), XGBoost performs better than GB and RF. Optimizing the hyperparameters of the XGBoost model significantly improves its performance. The tweaked XGBoost model shows a significant improvement in R2 (more than 5% to 10%) and error results (reduced MAE and RMSE by 0.01 to 0.06), when compared to other ensemble approaches. Compared to other ensemble approaches, the tuned XGBoost methodology is more robust and generates more accurate forecasts in solar power.