Traffic State Identification Based On HMM And Its Application In Traffic Flow Parameters Short-time Prediction

Traffic state identification can provide real-time traffic operating information for both road users and traffic managers.It plays an important role in road traffic information service and traffic management and control.This paper proposes a traffic state identification method based on Hidden Markov Model(HMM)using freeway section traffic data on California in America.In addition,based on the traffic state identification results from above,Random Forest(RF)for flow rate,speed,occupancy short-time perdition was constructed.Traffic states transition is a time-vary stochastic process and the relationship between unobserved traffic states and observed traffic flow is stochastic.Therefore,this paper models traffic states transition using HMM.Based on literature investigation,the number of traffic states is determined as five and traffic flow,speed,occupancy are selected as observation variables.Traffic state identification model is constructed based on three classic problems in HMM using freeway data.Finally,compared with traffic state identification method of Highway Capacity Manual(HCM),the result is evaluated from the aspects of traffic parameters scatter diagram and traffic parameters vector diagram.The results are in line with traffic flow characteristics,statistic and traffic state identification method of HCM.Considering about traffic states transition,Random Forest was chosen to predict flow rate,speed,and occupancy in short time.Traffic states from traffic state identification based on HMM were chosen as principal features as the variables to form input space.The importance of feature variables especially traffic state are analyzed based on Gini coefficient.Then the method proposed by this paper was compared with classic traffic flow parameters forecasting model Autoregressive Integrated Moving Average Model(ARIMA)which does not consider about traffic states.Results show that the general forecasting performance of traffic flow parameters forecasting model based on different traffic states is better that that of ARIMA.