Real-time Computing Of Road Traffic Volume Based On Streaming Taxi Trajectory

Real-time urban road traffic status based on the traffic volume is an important means to master traffic conditions.Taxi trajectory data is easy to obtain,covers a wide range,and up-to-date.It is an important data source for computing the traffic volume of urban road.Most of the research based on taxi trajectory is oriented to historical trajectory data set,processing trajectory data in batches and mining the traffic information currently.Due to the lack of real-time data,there is a serious lag in information,and it is difficult to effectively support time-sensitive traffic applications such as congestion evacuation and traffic scheduling.Aiming at the above problems,we have shown a real-time computing model of road traffic volume with high throughput and low delay based on the real-time streaming data.This thesis achieved the real-time road traffic status identification based on streaming trajectory.The following are our main accomplishments:1、A streaming map matching algorithm is proposed.We analyze the feasibility of map matching algorithm parallelization based on static Hidden Markov Model map matching;In addition,this thesis puts forward streaming map matching algorithm and achieves it taking stateful operation into account,which is first implemented on Apache Flink.2、Real-time traffic volume computing based on streaming taxi trajectory.Based on streaming trajectory which has been matched,we further focus on the real-time computing of road traffic volume.Firstly,we analysis the contribution of trajectory data to the road traffic status whether a taxi carries passengers or not.The study estimates the delay in the intersection area and considers the impact of the intersection range on the cross-section trajectory stream.Then the parallel computing strategy of the average travel speed of the road segment is studied.Finally,the real-time traffic status index(RTSI)suitable for the characteristics of urban road network and its classification are defined.This thesis shows the implementation process of the real-time computing model of road traffic status in Flink.3、Design and implementation of real-time computing system.The distributed message queue Kafka is select to achieve high concurrent trajectory stream injection.In order to facilitate the operation and computing of spatiotemporal streaming data,the original trajectory data was converted into JTS objects and serialized into binary byte streams based on Avro in the experiment.In the real-time computing module,Flink pulls the streaming trajectory from Kafka for consumption,and realizes each real-time computing operator by combining Flink programming model.Experiments are carried out oriented to streaming taxi trajectory.Streaming map matching is dynamic,matching snapshots are considered in this thesis.The matching results in different types of road network structures illustrate the effectiveness of the streaming map matching algorithm proposed in this thesis.We select 5 typical road segments to make a detailed spatial and temporal analysis of the real-time traffic volume and road traffic status in different periods,such as weekdays,weekends,morning peak hours,noon hours and evening peak hours.Based on the distributed cluster,the throughput of the system is nearly 2.7 times higher than that of a single machine and the acceleration ratio is about 2.19.under the corresponding parallelism setting,the real-time computing module throughput reaches 800,000 records per 10 min.The computing system has a high computational performance and can solve most largescale trajectory stream computing scenarios according to experimental results,such as real-time map matching,traffic volume computing and road traffic status acquisition.