基于Spark Streaming的海量GPS数据实时地图匹配算法  

Massive GPS data real-time map matching algorithm based on Spark Streaming

作　　者：陈艳艳[1] 李四洋 张云超 Chen Yanyan;Li Siyang;Zhang Yunchao(Beijing Key Laboratory of Traffic Engineering,Beijing University of Technology,Beijing 100124,China)

机构地区：[1]北京工业大学交通工程北京市重点实验室,北京100124

出　　处：《计算机应用研究》2024年第5期1338-1342,共5页Application Research of Computers

基　　金：国家重点研发计划资助项目(2022YFB2602104)。

摘　　要：浮动车GPS数据作为交通信息处理的基础,随着被监控车辆数量的高速增长,产生了海量GPS数据,对地图匹配提出了挑战。为了解决传统匹配方法难以满足匹配效率和精度的不足,提出一种针对海量GPS数据的实时并行地图匹配算法,能够同时保证较高匹配精度和运算效率。为构建一种面向实时数据流的高效、准确实时地图匹配算法,首先通过引入速度、方向综合权重因子对依赖历史轨迹的离线地图匹配算法进行重构,进而引入Spark Streaming分布式计算框架,实现地图匹配算法的实时、并行运算,大幅提升实时地图匹配效率。实验结果表明,该算法在复杂路段的匹配准确率较常规拓扑匹配算法提高10%以上,整体匹配准确率达到95%以上;在匹配效率方面,较同等数量的单机服务器效率可提高4倍左右。实验结果表明,该算法在由11台机器组成的计算集群上实现8000万个GPS数据点的实时地图匹配,证明了该算法可以完成城市地区的实时车辆匹配。Floating car GPS data serves as the foundation for processing traffic information,with the rapid increase in the number of monitored vehicles,a massive amount of GPS data is generated,posing great challenges to map matching.To address the shortcomings of traditional matching methods in terms of matching efficiency and accuracy,this paper proposed a real time parallel map matching algorithm for massive GPS data that ensured both high matching accuracy and computational efficiency.This paper firstly reconstructed an efficient and accurate real-time map matching algorithm for streaming data by introducing a comprehensive weight factor that considered velocity and direction to enhance the offline map matching algorithm that relied on historical trajectories.Then,it introduced the Spark Streaming distributed computing framework to achieve real-time and parallel computation of the map matching algorithm,significantly improving the efficiency of real-time map matching.Experimental results demonstrate that the proposed algorithm achieves more than a 10%increase in matching accuracy compared to conventional topological matching algorithms on complex road sections,with an overall matching accuracy of over 95%.In terms of matching efficiency,it achieves approximately a fourfold improvement compared to an equivalent number of standalone servers.The experimental results show that the proposed algorithm achieves real-time map matching of 80 million GPS data points on a computing cluster composed of 11 machines,proving that the proposed algorithm can achieve real-time vehicle matching in urban areas.

关 键 词：海量 GPS 并行计算 地图匹配 实时计算 SPARK 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]