Research On Ultra-Short-Term Wind Power Prediction Method Based On Wavelet Function And Time Series Theory

When wind power is connected to the grid on a large scale,the fluctuation and non-linearity of wind power bring huge potential safety hazards to the stable operation of the grid.Therefore,the ultra-short-term prediction of wind power can timely adjust and modify the grid operation scheme,thereby reducing the impact of large-scale wind power integration on the grid.In recent years,wind power ultra-short-term forecasting mainly adopts single or multiple forecasting methods combination method,but single forecasting method is liable to cause higher errors because of the defects of the forecasting model itself.Although the combined forecasting models of multiple forecasting methods can compensate for the defects of single forecasting model,they can not reduce the impact of non-stationary data itself.Therefore,based on the non-stationary wind power data,this paper proposes a prediction model which combines the wavelet function and time series method,and the main research contents are as follows.Research on power correlation of wind power.The data correlation analysis of wind power data and climatic factors affecting its change,including wind speed,air pressure,humidity,average temperature,maximum and minimum temperature,is carried out.Non-correlation data are eliminated through correlation analysis,which reduces a large number of useless experimental data,and also reduces the errors caused by data complexity in the later period.Selection of wind power prediction model.The exponential smoothing prediction model and the(Autoregressive Integrated Moving Average Model,ARIMA)in the non-stationary time series method are used to predict the historical data of wind power.The results show that although ARIMA prediction model can predict wind power effectively compared with exponential smoothing prediction model,the error will increase with the increase of prediction time.Therefore,it is necessary to determine an effective method for smoothing wind power.In this paper,four wavelet functions,Haar function,db N function,coif N function and sym N function,which are improved step by step on the same principle,are used to decompose the non-stationary irregular wind power data into a sum of data containing a large number of eigenvalues and data of N stationary and regular development trends.The results show that the combination of wavelet sym5 function and ARMA model can effectively improve the ultra-short-term wind power compared with the combination of wavelet Haar function and ARMA model,the combination of wavelet db3 function and ARMA model,and the combination of coif3 function and ARMA model.Precision of prediction.This method can reduce the impact on the grid when wind power is connected to the grid.It can also prepare the corresponding peak-shaving capacity to avoid unnecessary waste.