Forecasting of wind speed using wavelets analysis and cascade-correlation neural networks
Résumé
The present paper focuses on developing and testing various reliable and robust tools with the aim of managing energy sources and promoting renewable energy for the city of Perpignan (south of France). In this sense, forecasting average wind speeds was the main objective of the work, leading to propose a forecast methodology including a (discrete) wavelet-based multi- resolution analysis of available historical data, used as training data, and an estimation of the coefficients of the wavelet decomposition in the original resolution, using trained Artificial Neural Networks (ANN), of the next day, next week or next month average wind speed. Then, the reconstruction of a forecasted daily, weekly or monthly average wind speed is performed by simply summing up the estimated coefficients. According to the nature (that is to say daily, weekly or monthly) of a to-be-predicted average wind speed and because the purpose of the proposed methodology is identifying patterns in time series data, only previous days, weeks or months average wind speeds were considered. Finally, and because of the accuracy of the predicted average wind speeds, one can consider that the obtained results by means of the proposed DWT-ANN methodology are satisfactory even very satisfactory.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...