Random forest based wind power prediction method for sustainable energy system

Abstract

Wind power generation prediction is critical for the effective integration of renewable energy into the power grid, supporting stability, reliability, and sustainability in electricity supply. However, the inherent variability and non-linear characteristics of wind patterns present substantial challenges to accurate prediction. This study tackles these challenges by utilizing the Random Forest (RF) algorithm, an ensemble learning approach renowned for its ability to capture complex, non-linear relationships in data. The RF model’s performance is compared with three commonly used prediction techniques: Neural Networks (NN), Extreme Gradient Boosting (XGBoost), and Linear Regression (LR). The models were evaluated using historical wind power data and key meteorological variables, with performance assessed through multiple metrics, including Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Maximum Error (MAX), Standard Deviation (STD DEV), and R-squared (R²). The results indicate that the RF model achieved the best performance, with an RMSE of 55.11 and an R² of 0.9882, outperforming the NN, XGBoost, and LR models. Specifically, the NN model achieved an RMSE of 95.5 with an R² of 0.9651, XGBoost had an RMSE of 93.32 and an R² of 0.9666, and the LR model exhibited an RMSE of 144.45 with an R² of 0.9084. These findings demonstrate RF's superior predictive accuracy and robustness, making it a powerful tool for wind power forecasting, providing valuable insights for grid management and renewable energy planning.