A Hybrid Method based on Discrete Wavelets and Least Squares Support Vector Machines for Short-Term Wind Speed Forecasting

Exponential increase in power consumption leads to the global attention towards pollution free and renewable energy resources. For instance, wind turbines to produce electrical energy thru wind energy. For wind energy domain, wind speed forecasting is of great significance for wind farms design and planning, its operational control, and wind power prediction etc. Due to the impact of several environmental factors, time series of wind speed exhibits high fluctuations, less correlation and stochastic volatility. In this paper, to enhance the wind speed forecasting, a two stage method that relies on discrete wavelet (a trou’s wavelet transform) and least squares support vector machines (LS-SVM) is developed. In the first stage of the method, the time series of wind speed is decomposed into wavelet components by employing a trou’s wavelet. In the later stage, an LS-SVM is trained with the obtained wavelet components as inputs and h-samples ahead data as the target data (output). To identify the optimal initialization for LS-SVM, a grid search is conducted for a wide range of values. The nonlinear mapping obtained with the training data set and optimal initialization is employed to perform forecasting. The forecasting results obtained with the proposed method show better performance as compared to the existing methods.