Travel Time Estimation For Urban Road Networks Using ANPR Data

Travel time can indicate traffic state directly and timely,so many traffic engineers and researchers have a great interest in studies on travel time.Accurate,real-time and reliable travel time information,including link travel time and route travel time,is essential to support the Advanced Traffic Management Systems(ATMS)and the Advanced Traveler Information Systems(ATIS).However,travel time is unknown and uncertain,due to fluctuations in traffic demand and supply(e.g.due to the weather conditions and road geometry),traffic control(e.g.signal timing and lane groups),stochastic arrivals and departures at signalized intersections,etc.The traditional estimation methods based on traffic flow theory may lead a large error and the result cannot reflect the real traffic condition.Thanks to the well-established technologies,there are a verity of new data sources providing travel time information,like Floating Car Data(FCD),Automatic Vehicle Identified(AVI)data,Automatic Number Plate Recognition(ANPR)data,Floating Cellular Data(FCD)and Bluetooth data.Among of them,ANPR data based on the HD gate system contains multiple information of all passing vehicles,including plate number,the moment,entrance ID,departure lane ID.Traffic volume,travel times and driving directions of individual vehicles can be obtained.Meanwhile,the HD gate system is widely available in China.Therefore,we make a study on urban road travel time using ANPR data based on the HD gate system.Firstly,data analysis is conducted for ANPR data collected in a real urban network.We introduce the working principle,location,detected data and performance indicators of the HD gate system,before the section data analysis(including volume accuracy and recognition accuracy)and link travel time dada analysis.The relationship between the estimated parameters of travel time and the sampling rate is analyzed using the travel time data on road without openings.It is indicated that the Mean Absolute Percentage Error(MAPE)of estimated travel time and the fluctuation of standard deviation decrease as the sample rate increase;and the accuracy and stability is high when the sampling rate larger than 0.414,the threshold value.The match rate of travel time on the road with openings is computed,and its spatial-temporal characteristics and significant difference is analyzed.In the numerical analysis,the match rate of travel time has nothing to do with the link attribute and date,and it is always larger than the sampling rate threshold;the match rate of travel time is related with time periods,and it is lowest during 20:00-6:00 but still larger than the sampling rate threshold.We conclude that estimation result of travel time under good weather conditions can be good using ANPR data based on HD gate system.Secondly,we develop a method of travel time estimation based on traffic stream direction.Traffic flow on a given link in an urban road network can be divided into several traffic streams,depending on their turning movement when entering and leaving the link.These traffic streams may experience various travel times due to multiple reasons.However,the current travel time estimation methods take traffic flow as a whole and produce a single estimation value.This approach can produce large errors.In this paper,a comparison analysis is conducted to verify the significant difference in link travel times of different traffic streams.Then,link travel time is redefined in consideration of traffic stream directions.In addition,existing estimation methods cannot reflect real values or fluctuations of travel times in sampling intervals without any valid observational data.To solve this problem,a regression model is built and integrated into the travel time estimation model.Numerical experiments and error analysis of several links demonstrate the improvements in filtering noise and estimation accuracy made by the model.Finally,an estimation model is presented based on fusion of kinds of route travel time information.Route travel time obviously varies with vehicles and traffic demand.Therefore,in addition to average travel time value,travel time reliability on a route in the form of travel time distribution is indispensable.A route is re-identified using traffic streams,so observations grouping can exclude some non-representative travel time data.The paths of vehicles with small information gap are inferred,while others are split.Travel times of paths with small gap are scaled,while others are not.Route travel time data then is composed of complete route travel time(TTC)and partial travel time(TTP).Empirical distribution of TTC is taken as the estimated route travel time distribution,when the ratio of them is high.The route is divided into several parts by breakpoints connecting two consecutive links with weak correlation on the basis of Hopkins statistics.Convolution of empirical distributions of these parts using TTP is fused with TTC to estimate the real route travel time distribution,when the ration of TTC is low.256 cases in real network and simulation network are conducted to evaluate the effectiveness of the developed model,via the error analysis of estimation results,the impact analysis of recognition accuracy,the impact analysis of parameters,the impact analysis of route attributes and the comparisons of route travel time estimation results.