Study On Integrated Models Of Short-term Wind Speed Forecasting Based On Wavelet Transform

Wind power, one member of the clearer and pollution-free new energy family,has not only received wide attention but also seen a rapid development all over theworld. And recent years have seen a increasing capacity of wind power penetration.These moves are of great significance to bring down fossil energy consumption andreduce emission and pollution. However, the high uncertainties and fluctuation ofwind power have threatened the safe and economic operation of power system. Inorder to handle these, accurate forecasting of wind power is necessary which is basedon precise wind speed forecasting.For the non-stationary, non-linear and strongly fluctuating nature, wind speedsequences have been pre-processed with wavelet transform, a wonderful tool fortime-frequency localization and multi-resolution characteristics analysis. With thehelp of wavelet transform, it is possible of conducting gradually refiningtime-frequency processing of wind information elements, smoothing original windtime sequences and revealing characteristic information of historical data.After acquiring general wind and detail wind of characterized wind sequencesinternal characteristics of different frequency, a short-term wind speed forecastingmethod integrating autoregressive moving average model, autoregressiveheteroscedasticity model and general autoregressive heteroscedasticity model wasestablished. In the first place, corresponding autoregressive moving average modelscan be established according to various characteristics of each subsequence. Becauseclassical auto regressive moving average models (ARMA) are generally establishedbased on the same variance assumption which generally pay attention on time seriesof the first moment but rarely the second, the trouble of heteroscedasticity hides in theabsence of further study of residual series. Through autocorrelation Lagrangemultiplier tests （LM tests） of the residual series of each wind speed sequence`smaster model, it can be found out whether there is intact useful information in eachsubsequence. And ones with obvious test result are be modified by auto regressiveheteroscedasticity model and general auto regressive heteroscedasticity model to buildcorresponding ARMA-ARCH and ARMA-GARCH wind speed comprehensiveforecasting model for each subsequence. Through testing the average wind speedprediction in one hour of measured wind speed sequences, the method put forward in this paper are verified to effectively raise the forecast accuracy and fit better tothe.varying principles of wind velocity.