Abstract
•Research focuses on forecasting wind speed for wind energy conversion systems (WECS) planning.•Data collected from windy mountain city weather stations over 5 years.•A kernel ridge regression (RR) model is proposed and compared to SVM and ANN.•RR model showed best results in RMSE and R2 assessments.•Study provides valuable insights for cost and risk management in wind power planning.
For wind energy conversion systems (WECS), forecasting wind speed is crucial for meeting customer demands while monitoring, controlling, planning, and dispatching the electricity production. The goal of the research is to more easily predict wind speed for planning and feasible studies of wind farms. All data was derived from weather stations in a windy city in a mountain area for one month within five years. A kernel ridge regression (RR) model is suggested, and the results are compared with two reference prediction models of support vector machines (SVM) and artificial neural networks (ANN), to validate the model's efficiency for three different predicting horizons (1-h, 12-h, and 24-h ahead). The root means square error (RMSE), and root mean square (R2) are utilized to assess the effectiveness of a prediction model. Using one layer and 30 neurons, the optimum outcome was achieved with an RMSE of 0.264 and a value of 0.811. Tests revealed that 70%–30% for training and testing yields the lowest RMSE of 1.244 compared to 1.874 for 60%–40%. A study of ANN, SVM, and ridge regression found that predictions made with the RR provided the most precision in comparison to the R2 and RMSE values. This study's relevance lies in its ability to forecast wind speeds and for this reason, ridge regression using mutual information feature selection performs better than other methods when trying to forecast wind velocity. Cost and risk management in wind power planning will benefit from this research.