A New Hybrid Method for Wind Power Forecasting Based on Wavelet Decomposition and Artificial Neural Networks

Abstract

Depending on their input, wind power forecasting models are classified as physical or statistical approaches or a combination of both. Physical models use physical considerations, as meteorological information (Numerical Weather Prediction) and technical characteristics of the wind turbines (hub height, power curve, thrust coefficient). Statistical models use explanatory variables and online measurements, usually employing recursive techniques, like recursive least squares or artificial neural networks (ANNs) which perform a non-linear mapping and provide a robust approach for wind prediction. In this paper a new hybrid method (mixing physical and statistical approaches) is proposed, based on the wavelet decomposition technique and on artificial neural networks, in order to predict power production of a wind farm in different time horizons: 1, 3, 6, 12 and 24 hours. In particular, two approaches are compared, both based on the time series of on-line measured wind power and on the Numerical Weather Predictions; in the first approach, the forecast is carried out only through the training of a neural network which, in the second approach is, instead, used in combination with the wavelet decomposition technique, improving the performance especially over the short time horizons. The error of the different forecast systems is investigated for various forecasting horizons and statistical distributions of the error are calculated and presented.