Urban District Congestion Pattern Prediction Based On Hidden Markov Model

According to the evolution of urban traffic congestion,it often starts from intersection or road congestion and spreads to congestion in areas.Seriously,it will cause traffic congestion in large areas of the city.With the acceleration of urban motorization,the scope and duration of traffic congestion continue to broaden and prolong,and the phenomenon of congestion in district during peak hours is already common.Congestion in the district has become a normality in many large and medium-sized cities during peak hours,and it should be paid more attention to by the transportation industry management department.To deal with the congestion in urban areas,Objective prediction is the effective measure.Based on the taxi trajectory data,this paper proposes a method for predicting the congestion of district based on the Hidden Markov Model(HMM)for the road network area where three or more roads are simultaneously congested,mainly by revealing the evolution mechanism of traffic flow.and it makes full use of the spatiotemporal correlation between the peripheral and internal traffic states to accomplish prediction.The method is different from the commonly used extrapolation method,and is expected to improve the prediction accuracy of congestion patterns in the district.The steps and corresponding finding of this study are shown as following:(1)Firstly,this paper cleans the original taxi historical trajectory data and part of the Gaode-Map data to obtain valid trajectory data.(2)Secondly,by matching with the vector map of the urban road network,the dense road segment set of taxi trajectory data is extracted as a candidate for the congested area,and the minimum discrete time window is repeatedly tested to ensure the continuous gradual change of the dynamic vehicle speed,and finally the speed sequence of the road segment in different time window is conformed.(3)Thirdly,based on the road speed sequence and the traffic congestion degree of the road sections,the traffic state of each road section is determined,and the traffic congestion patterns of the road sections are integrated and identified,which is used as the hidden states set of the prediction model.(4)Fourthly,using the equal-time traffic accessibility theory to trace the upstream bayonet in the periphery of the area,and discretize the road speed of the upstream bayonet as the observation states set of the prediction model.(5)Fifthly,a prediction model of the congestion pattern in the area is established.Using historical taxi trajectory data to train the model and determine model parameters.The real-time Gaode-Map data is used to predict the congestion pattern of the area,and this model compares with another commonly used time series model: Autoregressive Moving Average Model(ARIMA)to verify the prediction reliability of the model.In this paper,considering the traffic interaction between road sections,the upstream traffic is used to predict the future traffic in the downstream area,and the case of Lihuili hospital in Ningbo is taken as an example.This paper analyzes the hidden state transition probability matrix parameters of the model and the observed state confusion matrix parameters to find the propagation rules of the congestion patterns in the area.It is useful to predict the congestion change in the area and help the traffic control center to clear the congestion in advance.At the same time,the traffic broadcast will be released to help travelers choose their way of travel in advance and avoid congestion.