基于马尔科夫模型的出行目的地预测

发布时间：2018-01-19 19:33

本文关键词： 出行目的地 GPS 频繁到访目的地支持点马尔科夫模型　出处：《吉林大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来随着手机导航软件和GPS(全球卫星定位系统)设备的广泛使用,GPS数据成为分析居民出行行为的新途径。通过对GPS数据进行挖掘,提取出行信息用于出行目的地预测是出行行为分析研究中的一个比较复杂的过程。进行出行目的地预测,准确地分析居民出行空间分布特征,据此对城市交通系统进行合理的规划和布局,是改善交通问题的有效方法,同时对于保障城市交通的可持续发展具有深远意义。本文根据对GPS数据的处理分析,提出了基于马尔科夫模型的出行目的地预测模型体系。首先,对GPS出行数据进行调查处理,将GPS设备调查得到的GPS轨迹点转化为连续的出行段,确定出行目的地。并划分工作日和非工作日的不同出行,设定合理的参数从中提取并确定频繁到访目的地和支持点信息,获得建模数据。其次,分别基于马尔科夫链模型和隐马尔科夫模型进行出行前和出行中的目的地预测。在出行前和出行中,对工作日和非工作日的建模数据分别进行目的地预测模型建立和标定,利用居民出行GPS数据进行实例分析。结果表明,应用已建的出行前和出行中的目的地预测模型均能够获得较高的预测精度。相比较而言,出行中的目的地预测因为有支持点加入,预测效果更好。且由于工作日和非工作日的划分,与以往研究相比预测精度更加稳定。因此本文提出的出行目的地预测方法可用于居民出行目的地预测,既可以在出行前进行出行目的地的预判,又可以在出行中进行实时的目的地修正。本文提出的出行目的地预测的马尔科夫模型无需进行传统的入户式居民出行调查,无需道路地图,减少了数据调查的复杂性,简化了数据处理的过程。研究成果可应用于车载导航设备或手机导航软件中,通过采集出行者的GPS数据,在出行前或出行中进行实时的出行目的地预测,以节省出行者输入目的地的操作步骤,提高导航运算速度和行车安全性。并可以向出行者推荐出行路线及出行目的地附近的加油站、商场等设施,满足出行需求。同时,出行目的地预测结果还可以在交通规划和管理方面广泛应用,比如居民出行空间分布特征分析、出行分布预测、拥挤位置预测分析等。再者,研究居民出行目的地选择的非集计行为模型对于促进交通需求预测从传统的集计四阶段模型向非集计行为模型的转变具有一定意义。
[Abstract]:In recent years, with the wide use of mobile phone navigation software and GPS (Global Positioning system) equipment, GPS data become a new way to analyze residents' travel behavior. It is a complicated process to extract travel information for travel destination prediction in the research of trip behavior analysis. The travel destination prediction is carried out and the distribution characteristics of residents' travel space are analyzed accurately. Therefore, it is an effective method to improve the traffic problem by rational planning and layout of urban traffic system. At the same time, it is of great significance to ensure the sustainable development of urban traffic. Based on the analysis of GPS data, this paper puts forward a new model system of trip destination prediction based on Markov model. The GPS travel data are investigated and processed, the GPS locus obtained from the GPS equipment survey is transformed into continuous travel segments, and the travel destination is determined. The different trips are divided into working days and non-working days. Set reasonable parameters from which to extract and determine frequent visit to the destination and support point information to obtain modeling data. Based on Markov chain model and hidden Markov model to predict the pre-trip and in-trip destination respectively. The modeling data of working day and non-working day are established and calibrated respectively, and the GPS data of residents travel are used to analyze the example. The results show that. Using the established pre-trip and in-trip destination prediction models can achieve a higher prediction accuracy. In comparison, the destination prediction in travel is due to the addition of support points. Because of the division of working day and non-working day, the prediction accuracy is more stable than that of previous research. Therefore, the method proposed in this paper can be used to forecast the travel destination of residents. It is possible to predict the destination of the trip before the trip. The Markov model proposed in this paper does not need to carry out the traditional household residents travel survey, no road map. It reduces the complexity of data survey and simplifies the process of data processing. The research results can be applied to vehicle navigation equipment or mobile phone navigation software by collecting the GPS data of travelers. The real time travel destination prediction is carried out before or during the trip to save the traveler from entering the operation steps of the destination. Improve the speed of navigation and driving safety. And can recommend the route and travel destination near the gas station, shopping malls and other facilities, to meet the travel needs. At the same time. Travel destination prediction results can also be widely used in traffic planning and management, such as residents travel space distribution characteristics analysis, travel distribution prediction, congestion location prediction analysis, and so on. The study of disaggregate behavior model for residents' travel destination selection is significant to promote the change of traffic demand prediction from the traditional four-stage model to the disaggregate behavior model.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O211.62;P228.4

【相似文献】