Short-term wind speed prediction based on LMD and improved FA optimized combined kernel function LSSVM

Abstract

Accurate prediction of wind speed is of great significance to the operation and maintenance of wind farms, the optimal scheduling of turbines, and the safe and stable operation of power grids. A new prediction method for short-term wind speed based on local mean decomposition (LMD) and combined kernel function least squares support vector machine (LSSVM) is proposed. The short-term wind speed time series is decomposed into some components by the LMD algorithm. Based on LSSVM, radial basis function and the Polynomial function are used to generate the combined kernel function. The combined kernel function LSSVM combines the advantages of the radial basis function and the Polynomial function, which can achieve better prediction accuracy. The decomposed wind speed time series are predicted separately by the combined kernel function LSSVM model. At the same time, an improved firefly algorithm is proposed to optimize the parameters of the combined kernel function LSSVM. The final predictive value can be obtained by superimposing the predicted value of each combined kernel function LSSVM prediction model. The actual collected short-term wind speed data is chosen as the research object, the simulation experiments with four prediction horizons have been implemented. Compared with state-of-the-art prediction methods, through the comparison result curve between the prediction and actual wind speed, the box-plot results of predictive error distribution, the comparison results of the relative prediction error, the performance indicators, the Pearson’s test, the DM test and the Taylor diagram results show that the proposed prediction method has higher prediction accuracy and is able to reflect the laws of wind speed correctly. Furthermore, the simulation results of four new datasets and adding noise to the input data of training set show that the proposed prediction method has strong robustness.