Parallel Map Matching Method For Large-Scale GPS Trajectory Data

Trajectory data mining is a current research hotspot in the field of intelligent transportation.Research on path navigation and traffic behavior analysis requires trajectory data containing road segment information.However,The GPS trajectory data collected by the taxi GPS device has no road segment information of the vehicle,and the latitude and longitude coordinates of the trajectory point have errors.The map matching algorithm can match the road segment to which the trajectory point belongs in the road network data,thereby correcting the latitude and longitude coordinates of the trajectory point.Therefore,map matching method research is a necessary basic research content in trajectory data mining.The traditional map matching method for GPS trajectory data usually mainly considers the accuracy of the matching results,and ignores the matching efficiency.With the rapidly growth of the number of vehicles,the trajectory data shows an explosive growth trend.In this situation,the traditional map matching method is inefficient,and the map matching efficiency problem of large-scale trajectory data needs to be solved urgently.Therefore,this paper proposes a parallel map matching method for large-scale GPS trajectory data.The main contents of this paper are as follows:(1)Most of the map matching algorithms determine whether the road segment belongs to the current candidate road segment set according to whether the projection distance of the current trajectory point in each road segment is less than a preset distance threshold.In the face of large-scale trajectory data sets and large-scale road networks,the above-mentioned candidate road segment set selection methods are relatively inefficient.A distributed grid map indexing method based on GeoHash coding is proposed,which can effectively improve the efficiency of candidate segment selection.(2)In the face of massive vehicle trajectory data,the computational efficiency of the traditional map matching algorithm has been unable to meet the needs of related research work.To improve this problem,a parallel map matching method is proposed.A time-based partitioning strategy is proposed to effectively improve the data skew problem in parallel map matching.The experimental results show that the proposed method can achieve a matching throughput rate of 85,400/s under the condition of high accuracy.Compared with a parallel map matching algorithm based on Hadoop proposed in the literature,its operation speed is increased by about 33 times.(3)The parallel map matching algorithm proposed in the research content 2 can not solve the map matching problem of real-time trajectory data stream.Based on the Structured Streaming computing model,an online map matching method for large-scale trajectory data is further proposed,which guarantees a certain real-time.Under the condition of ensuring certain real-time performance,the stream processing of map matching is realized.(4)A method for recommending taxi passengers waiting segments based on ensemble learning is proposed.First,the number of taxis on the various segments of Xi'an in a certain time interval is counted.Then,the ensemble learning method is used to predict the number of empty taxis in the next time slot of each segment of Xi'an.Finally,based on the prediction results,the best waiting segment is recommended for the upcoming passengers.