蒋鸿玲　张　楠　李　克　田　昊　葛　伟

(中国航天科技集团公司物联网技术应用研究院　北京 100094)



基于MapReduce的出租车停泊点智能推荐算法

蒋鸿玲张楠李克田昊葛伟

(中国航天科技集团公司物联网技术应用研究院北京 100094)

摘要现有解决打车难问题的研究工作大部分是集中式地调度出租车，且大多方法在单一服务器上运行串行算法分析海量出租车GPS数据，计算量大，会遇到计算时间和计算资源的瓶颈。为此提出一种基于MapReduce的出租车停泊点智能推荐算法，为司机或乘客推荐更容易接到乘客或打到车的地点。算法通过挖掘大量出租车GPS轨迹数据，检测出停泊点，并生成停泊点知识库。再利用推荐模型为司机或乘客推荐最佳停泊点。实验分析了北京市真实出租车GPS轨迹数据，结果表明该算法能有效为司机和乘客推荐出停泊点，且在大数据量下具有较高的效率。

关键词出租车GPS聚类MapReduce停泊点推荐

TAXI PARKING POINT RECOMMENDATION ALGORITHM BASED ON MAPREDUCE

Jiang HonglingZhang NanLi KeTian HaoGe Wei

(Internet of Things Technology Application Institute, China Aerospace Science and Technology Corporation, Beijing 100094,China)

AbstractMost researches on solving the problem of hard to take a taxi are for dispatching taxi centrally. The majority of approaches have huge computation and may encounter the bottlenecks of computation time and calculation resources, because they need to analyse a large amount of GPS data on a single server operating the serial algorithm. So the paper proposes a MapReduce-based taxi parking point intelligence recommendation algorithm. It recommends for drivers and passengers the places where the passengers or taxis are more likely to be met or taken. By mining a large amount of taxi GPS trace data, the algorithm detects the parking points and generates the knowledge base of parking points. Then it uses recommendation models to recommend the best parking points for drivers and passengers. The experiments analysed actual taxi GPS trace data in Beijing. Results showed that the algorithm could effectively recommend parking points for drivers and passengers, and had higher efficiency in large amount of data.

KeywordsTaxiGPSClusterMapReduceParking pointRecommendation

0引言

目前打车难成为城市交通的普遍问题。乘客常打不到车，而很多出租车在路上盲目空载行驶，不仅浪费时间给城市交通带来额外的负担，而且浪费能源，影响城市的环境。现有一些研究关注出租车调度系统和出租车GPS数据分析。文献[1]综述了现有出租车GPS数据分析的研究方向，包括分析城市人群流动情况、分析不同地区和时段的交通拥堵情况、为乘客和出租车司机提供服务等。文献[2]调度空车去乘客数期望最高的地点。文献[3]用概率模型分析司机和乘客选择不同策略时的成本及风险，并为司机或乘客推荐停泊点及路径。……