Research On Short-term Traffic Flow Prediction At Urban Intersection Based On ARIMA-MCC And CKDE-GARCH

Short-term traffic flow prediction has become one of the hotspots in the field of transportation research.It can provide real-time and reliable information for traffic management departments,and then make corresponding decisions to alleviate traffic congestion.Due to the existence of linear and heteroscedastic components in short-term traffic flow,the researchers proposed the autoregressive integrated moving average(ARIMA)model and the Generalized Autoregressive Conditional Heteroskedasticity(GARCH)model to explain the linear components and the heteroscedastic of the residuals.However,the ARIMA model needs to meet basic assumptions,the residuals of this model must meet the Gaussian distribution,which is difficult to meet in practical applications.In addition,the residuals estimated by the traditional GARCH model are obtained through standard random variables,which may produce inaccurate estimation and thus cannot explain the heteroscedasticity of the residuals well.Aiming at the shortcomings of the traditional short-term traffic flow hybrid model(ARIMA-GARCH),this paper proposes a new model.The new hybrid model is divided into two parts.The first one is the improved ARIMA model.In this paper,the maximum correlation entropy criterion(MCC)is used to estimate the autocorrelation coefficient and partial autocorrelation coefficient of the ARIMA model;The second one is the improved GARCH model,that is,the nonparametric model-Conditional Kernel Density Model(CKDE)is adopted to estimate the residuals of the ARIMA-MCC model,and the estimated value is utilized to replace the random variable of the conventional GARCH model.the coefficients of the CKDE-GARCH model are obtained by fitting the residuals estimated by the CKDE-GARCH model and the residuals of the ARIMA-MCC model.In the end,the step ahead residuals are estimated by the CKDE-GARCH model to compensate the predicted values of ARIMA-MCC.Finally,the new hybrid model is used to predict the short-term traffic flow of the real urban intersection,and the corresponding prediction results are analyzed.In order to reveal the superiority of the new model,some targeted comparison models are selected to compare with the proposed model.In addition,this paper also selects another three groups of short-term traffic flow with different characteristics to verify the proposed model,and analyzes the applicability of the new hybrid model.