Wind power forecasting based on new hybrid model with TCN residual modification

Abstract

Wind energy has been widely utilized to alleviate the shortage of fossil resources. When wind power is integrated into the power grid on a large scale, the power grid's stability is severely harmed due to the fluctuating and intermittent properties of wind speed. Accurate wind power forecasts help to formulate good operational strategies for wind farms. A short-term wind power forecasting method based on new hybrid model is proposed to increase the accuracy of wind power forecast. Firstly, wind power time series are separated using the complete ensemble empirical mode decomposition with adaptive noise method to obtain multiple components, which are then predicted using a support vector regression machine model optimized through using the grid search and cross validation (GridSearchCV) algorithm. Secondly, a residual modification model based on temporal convolutional network is constructed, and variables with high correlation are selected as the input features of the model to predict the residuals of wind power. Finally, the prediction accuracy of the proposed method is compared to other models using the actual wind power data of the wind farm to demonstrate the validity of the described method, and the results reveal that the proposed method has better prediction performance.© 2017 Elsevier Inc. All rights reserved.