The Research On Wind Speed Prediction For WTG Based On ARMA Model

With the currently rapid development of wind power technology, large-scale wind power plant has began gained access to power grid, however, with certain potential hazards on operating safety and stability. For this matter, in order to coordinate with power administrations and ensure the safe operation of power grid, high precision of the wind generator power forecast, which is based on accurate predictiona of wind speed, is very important.Based on time series prediction method, this paper introduces basic principles of wind speed prediction and those of time series model, and specifically based on ARMA model, studies wind speed of multi-point prediction.First of all, this paper runs simulations and verifications of ARMA time series prediction method, using field data of wind power acuqired from a domestic wind field. Then according to the analysis of the possibility of one-time muliti-point prediciton, a variavle step ARMA predition, which enables multi-point prediction ensuing from a future time point by a series of constant historic data(those multiple points donnot have to exactly follow it), is presented.Secondly, this paper also studies whether there can be furthermore forecasted period, and presents forecasting method of wind speed signal trend decomposition. Taking the fact that wind speed will not vary significantly in long term as advantage, former wind signal is divided into trend signal and de-trend signal, then ARMA prediciting model is formulated to get following preditions.Because this method is extremely particular about the data selection, wavelet decomposition and sliding filtering are introduced before modeling of ARMA as pre-processing to contain unnecessary error and in turn ensure the precision. To fuether enhance its precision, the deep fusion of wavelet decomposition and sliding filtering is also conducted, presented as a combined ARMA predicting method. All the algrithms mentioned are realized on monitoring platform through embedded MATLAB nodes, and verified by field data.Finally, some conclusions are drawed, concerning research results and offering convenience for future research.