Publication

Abstract :

Accurate short-term wind power forecasting is crucial for reliable integration of renewable energy into the grid. This work offers a comparative analysis of statistical, machine learning, and deep learning models using multivariate time series data from an individual turbine. Five models are assessed: Seasonal Auto-Regressive Integrated Moving Average with Exogenous Variables (SARIMAX), Support Vector Regression (SVR), Random Forest, Extreme Gradient Boosting (XGBoost), and Long Short-Term Memory (LSTM) networks. Results indicate that ensemble tree-based models, especially Random Forest, achieve the highest accuracy (R2 =0.914), with SVR and SARIMAX also demonstrating competitive performance. XGBoost offers reasonable accuracy, while LSTM networks underperform due to insufficient temporal resolution in daily-aggregated data. The study emphasizes trade-off among accuracy, interpretability, and computational cost. The findings offer practical guidance on model selection for short-term forecasting under data constraints.