Abstract:
In order to capture the uncertainty of short-term traffic forecasting caused by the random fluctuation of traffic flow, the heteroscedasticity which can reflect the fluctuation is used to quantify the reliability of traffic forecasting. On the basis of time series and its heteroscedastic theory, a generalized autoregressive conditional heteroscedasticity (GARCH(1,1)) model was developed, in which an autoregressive integrated moving average (ARIMA(0,1,1)) model was used as the mean equation. The ARCH LM test results show that the heteroscedasticity of the ARIMA(0,1,1) model can be effectively captured and eliminated by the proposed GARCH(1,1) model. Performance evaluation illustrates that based on the GARCH(1,1) model, the traffic volume forecasting of urban expressway has a mean absolute percentage error (MAPE) of less than 10%, and the speed forecasting of urban expressway and arterial roads has a MAPE between 7.86% and 10.24%. Compared with the fixed confidence intervals predicted by ARIMA(0,1,1) model, the GARCH(1,1) model can produce narrower forecasting confidence intervals on the premise of effective prediction of free flow traffic conditions; while in congested traffic conditions, the GARCH(1,1) model can produce wider forecasting confidence intervals to improve the forecasting reliability by reducing the invalid prediction.