Application Of GRNN And Markov Combined Model Based On SVR Outlier Detection And VMD Decomposition In Wind Speed Forecasting

The quality of the model based on the quality of the collected data. However, due to various uncertainties and the impact of human negligence, the collected data exists abnormal values which will significantly affect the results of the model, so detecting the abnormal value is necessary in the first step. What is more, the decomposition of time series data is also a commonly used method of data preprocessing, the subseries can reflect the different characteristics of the original data, which can provide accurate information for further data analysis.In this paper, we take wind speed as the study object. On the one hand, wind energy is clean energy; on the other hand, as a sustainable energy, wind resource provides an adequate source for wind power, so the prediction of wind speed is particularly important.Based on the wind speed data of the Sotavento Galicia wind field in Spain, this paper uses the support vector machine regression(SVR) method to detect the outliers of the original data. Then the variational mode decomposition(VMD) method is applied to decomposed the series which has removed the outliers. Then the generalized regression neural network(GRNN) is used to predict the subseries respectively. Finally, the final predicted vaule is obtained by sum up the prediction values of all subseries and adjusted with error correction by Markov chain. The experimental results show that the SVR outlier detection method, VMD decomposition method and Markov error correction model are useful for the prediction accuracy of the model. In addition, a comparison between the proposed model and other conventional methods is carried out to demonstrate its efficiency and estimation performance.