A novel approach to multi-horizon wind speed forecasting based on advanced machine learning techniques using sparse meteorological data of indigenous wind farm of Pakistan
TL;DR
The results validate that ensemble learning models are more reliable for sparse Meteorological datasets, whereas DL models are effective for capturing short-term temporal patterns, and provides a practical forecasting framework for wind farms, offering improved operational planning, reduced forecasting uncertainty.
Abstract
Wind energy represents one of the important sources of renewable energy (RE) and plays a pivotal role in the international decarbonization of energy systems. The novelty of the research work lies in the development and evaluation of a multi-horizon forecasting strategy under sparse-data conditions, where limited but evidently meaningful meteorological variables, that is, wind speed, wind direction, temperature, air pressure, and humidity, are used for short-term prediction, because these variables are important for accurate forecasting. Numerous Machine Learning (ML) and Deep Learning (DL) models, including Decision Tree (DT), Random Forest (RF), Support Vector Regression (SVR),Extreme Gradient Boosting (XGB), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU), the models were developed, and their performance was compared and analyzed for 10-minute, 30-minute, and 60-minute onward wind speed forecasting, and the results show that the Random Forest model achieved strong short-term forecasting performance at the 10-minute horizon with Mean absolute error (MAE) of 0.4364, Mean square error (MSE) of 0.3544, and Coefficient of determinations (R 2 ) of 0.9157, while the GRU model demonstrated modest temporal learning capability with R 2 values beyond 0.91 for short-term prediction. At the 30-minute horizon, GRU achieved the highest R 2 of 0.877, whereas RF maintained low prediction errors with MAE of 0.5628 and MSE of 0.5756, and for the 60-minute horizon, RF remained the most stable model, producing MAE of 0.6938, MSE of 0.8738, and R 2 of 0.7778, indicating better robustness than DL models under limited data conditions. The results validate that ensemble learning models are more reliable for sparse Meteorological datasets, whereas DL models are effective for capturing short-term temporal patterns, subsequently this research provides a practical forecasting framework for wind farms, offering improved operational planning, reduced forecasting uncertainty.