Research On Short-term Wind Speed Prediction Method Based On EWT And LSSVM

With the development of society and economy,the demand for energy is increasing.Non-renewable energy sources,such as coal,oil and natural gas are not only limited in quantity but also seriously polluting the environment.As non-polluting and renewable energy source,wind energy has become an important alternative to traditional fossil fuels.Accordingly,more and more attention has been paid to wind power generation.Wind speed is an important factor in wind power generation.However,wind energy is intermittent,unstable and other factors,making accurate prediction of wind speed is the key to solve the problem of wind power integration.Therefore,it is very important to design more accurate wind speed prediction model for the development of wind power generation.Least Squares Support Vector Machine(LSSVM)is an artificial intelligence prediction method,which has the advantages of simple calculation and strong robustness,and has a good effect on nonlinear data processing.Empirical Wavelet Transform(EWT)is a data preprocessing method that can denoise the original wind speed data.In view of the high noise and nonlinear characteristics of the original wind speed data,the paper studies the short-term wind speed prediction method based on EWT and LSSVM.A LSSVM model based on a four-parameter mixed kernel function is proposed.Firstly,empirical wavelet transform(EWT)is used to decompose the original wind speed signal,and an adaptive threshold function is designed to denoise each component.Secondly,a new model with mixed kernel function is designed to ensure the generalization ability and learning ability of the model.Finally,as the proposed model's four parameters of the proposed model(regularization parameters ?,RBF kernel width?,poly kernel parameters q,and mixed kernel weight coefficients c)have great influence on the prediction results and are difficult to be determined artificially,the cuckoo search(CS)algorithm is used to optimize the model's parameters,which improves the prediction accuracy of the model.Wind speed data not only contains non-linearity,but also has some non-stationary characteristics.LSSVM has poor adaptability to non-stationary signals.Therefore,this paper proposes a combined prediction model based on the four-parameter mixed kernel improved LSSVM model and differential autoregressive moving average(ARIMA).Firstly,the signal is decomposed into low frequency approximate component and high frequency detail components by EWT algorithm.For low frequency component,the four-parameter mixed kernel LSSVM model proposed in this paper is used for prediction.For high frequency components,ARIMA model with good applicability to non-stationary signals is used.Finally,the prediction results of each component are reconstructed to obtain the final prediction results.Through the signal decomposition and classification mixed prediction,the prediction accuracy is further improved.For the two wind speed prediction methods proposed in this paper.The results of simulation experiments show that the LSSVM model improved based on the four-parameter mixed kernel function has low algorithm complexity and is suitable for real-time wind speed prediction.However,the combination forecasting model based on the four-parameter mixed kernel improved LSSVM model and ARIMA greatly improves the forecasting accuracy,and is suitable for the situation where the forecasting accuracy is required to be high.