Study On The Model Of Short-time Wind Power Forecasting Based On Multi Scale Decomposition And Chaos Theory

With the further shortage of the globe resource, wind power is becoming more ofconcern as a clean, rich-reserving,and renewable energy. The most urgent problem is topromote the accuracy of wind power prediciton during the trend of connection betweenthe large wind farm and traditional power systems. Nowadays, the study of wind powerprediction pays more attention to the good or bad attribute of forecasting model,ignoring the analysis of the charactoristic of wind power time series,which results theprediction model cannot learn all the information about the wind power time series fully.According to what is metioned above,this thesis uses multi scale decomposition andchoas theory to modify extant wind power prediction model,which is demonstrated asfollows.This paper decomposes the wind power time series using the empiricaldecompositon theory, further, makes the Hilbert—Huang transformation for eachfrequency band component，gets the time—frequency spectrum and marginal spectrumof wind power time series，which reflects the change of amplitude of wind power timeseries signal according to frequency and time.This paper uses C—C method for phase space reconstruction，at the same time，estimating delay time,embedded demention and counting the largest Lyapunov index.So,we can use chaotic prediction model for them when we judge each frequency bandcomponent has the choatic characteristic.On the foundation of phase space reconstruction,this paper modify Least Square—Support Vector Machine and Radial Basis Function Neural Net for wind powerprediciton.According to the sample,these two model can finely reflect the tendecy of thechange of future wind power and gain very high predicting accuracy..For reducing theerror of prediction model, This thesis utilizes the Partical Swarm Algorithm foroptimizing the parameter of these two modelsConsidering both the regulation of every frequency band and the advantage ofindividual prediction model, this paper puts forward a longitude superposition windpower forecasting model:Firstly, caculating average instant frequency of each frequencyband signal,distinguishing high frequency randomness components and low frequencytendency components.Secondly, implementing the chaotic prediction using neuralnetwork with radical basis function for random componets, performing the chaotic prediction using least squares support vector machine for trend components.Thirdly,obtaining the final consequence by combining the prediction result of every component.The result indicates that the method of longitude superposition prediction is better thaneach single model, What is more, it can forecast the wind power more steadily andprovide reference to the study of wind power prediction.