Method of Short-Term Wind Speed Forecasting Based on Generalized Autoregressive Conditional Heteroscedasticity Model

In order to improve the security of the train operation, a short-term wind speed forecasting method was proposed based on a linear recursive auto-regressive integrated moving average (ARIMA) algorithm and a non-linear recursive generalized auto-regressive conditionally heteroscedastic (GARCH) algorithm (ARIMA-GARCH). Firstly, the non-stationarity embedded in the original wind speed data was pre-processed to eliminate its effect on the proposed model. Then, a linear recursive ARIMA model was build to predict wind speed. Finally, a non-linear recursive forecasting model was established based on the GARCH algorithm. Numerical example based on wind samples from field measurements The result shows that compared with the source data, the proposed approach has a higher prediction accuracy. With the increasing of prediction step, the mean absolute error only increases from 0.836 m/s to 1.272 m/s. The new method explains parts of the non-linear characteristics (heteroscedasticity) of wind speed time series and improves the prediction accuracy compared with the usual ARIMA approach. For instance, compared with the ARIMA model, the accuracy of mean absolute error of the six-step in advance forecast based on the GARCH model is improved by 11.54%.