With the deterioration of the global environment and reduction of the non-renewable en-ergy,people look forward to finding clean energy can replace non-renewable energy.As a kind of renewable clean energy,wind energy has brought to the attention of more and more countries.At present,the main application of wind energy in wind power from wind speed.However,there are some characteristics of randomness,instability and nonlinear in wind speed.In order to guarantee the security and economic benefits of power system,it is nece s-sary to understand the future short-term wind speed to make a response in advance.Wind speed sequence is a kind of time series,so wind speed forecasting equals time se-ries forecasting.At present,a lot of wind speed prediction models have been proposed,but there were some questions.The first,generally only forecasting wind speed and no reducing the error handling;The Second,forecasting algorithms proposed are not a very good solution to solve the characteristics of randomness,instability and nonlinear in wind speed forecasting,which results in the error of wind speed forecasting is large;The third,forecasting algorithms proposed are not good to keep the original wind speed sequence characteristics,parts of them not preprocess simply.In order to improve the accuracy of wind speed forecasting,this paper mainly propose two hybrid models of wind speed forecasting.The one is that a hybrid forecasting model based on extreme learning machines(ELM)and complete ensemble empirical mode decom-position(CEEMD)is proposed to realize the wind speed forecasting.First,non-stationary time series is decomposed into a series of stable components by using CEEMD.Then,the forecast of each component based on the ELM.Finally,compose the forecast of the compo-nents to get the final forecast.Another one is that the papers proposed a hybrid model of Complete Ensemble Empiri-cal Mode Decomposition(CEEMD),Wavelet Transform(WT)and Convolutional Neural Networks(CNN)to improve forecasting accuracy.First,CEEMD decomposed original wind speed into some relatively stable intrinsic mode functions and a residual sequence.Then,WT made secondary noise elimination to eliminate effects of noise on each intrinsic mode func-tion.Finally,the final result is obtained by refactoring forecasting results that CNN trained each intrinsic mode function,residual sequence and five attribute to obtain respectively.Use the random fuzzy uncertainty theory to secondary forecast for the two models of above.The paper proposes two wind speed forecasting model above from preprocessing of the original wind speed,traiing algorithms and processing of final results of forecasting. |