Wind speed and wind power forecasting using wavelet denoising-GMDH neural network

Abstract

Short-term forecasting tool for wind power generation (WPG) can effectively enhance the scheduling of dispatchable generation and avoid wind power (WP) curtailment. This paper presents a Group Method of Data Handling (GMDH)-Neural Network (NN) approach for forecasting wind speed (Ws) and wind power output (WPO) in the short-term. Wavelet denoising is used to filter the high frequency outliers in the Ws data, consequently, enabling a smooth training of NN on Ws. The performance of the wavelet denoising-GMDH-NN investigated using historical Ws and WPO. The results highlight that GMDH-NN performs well for Ws prediction and that the use of wavelet denoising with GMDH-NN results in a higher precision and lower errors, thus improving the prediction accuracy.