Real-Time Short-Term Traffic Flow Adaptive Forecasting Method Based On Time Series Analysis

Real-time and accurate short-term traffic flow forecasting has become a criticalproblem in intelligent transportation systems (ITS). The strong random disturbanceand strong uncertainty of model bring much difficulity for accurate traffic flowforecasting . In addition, nonstationarity of traffic flow data series is another problemto be solved. Based on time series analysis method adopting AR(p) model, a kind of real-timeadaptive forecasting method for short-term traffic flow was presented . In this methodthe recursive forgetting factor least square method (RFFLS) was adopted forparameter estimation. The Astrom forecasting algorithm was used for forecasting,which is based on linear minimum square error of prediction. A lot of real observationdata are used for simulation tests and results show that when forgetting factor isdecreased, the one-step forecasting performance can be improved. In addition, whenthis method is respectively applied to the data at the weekday and the weekend, bothsimulation tests have good forecasting performance, which demonstrates that thismethod has good adaptability in different traffic flow circumstances. Then based on this method a kind of improving multi-step adaptive forecastingmethod was presented. The error compensation item was added to this new methodwhich could well meet the needs of forecasting for time-variant models. A lot of realobservation data are used for simulation tests and results show that when theimproving method is applied to the strong time-variant multi-step short-term trafficflow forecasting, it has good forecasting performance which is superior to the linearminimum square error forecasting method's. Another adaptive forecasting method based on GM(1,1) model is also appliedto the short-term traffic flow forecasting and simulation results show it also has goodforecasting performance. But its forecasting performance is not superior to theadaptive forecasting method's based on AR(11) model. In order to utilizing advantages of these two kind of models, a kind of adaptiveforecasting method based on combination model of GM(1,1) model and AR(11)model was presented. simulation results show it has good forecasting performancewhich is superior to the forecasting method's based on the single GM(1,1) model orsingle AR(11) model and it is a kind of much better forecasting model.