基于GPS数据的公交车到达时间预测技术的研究

发布时间：2018-05-07 22:05

本文选题：实时公交 + 公交车到达时间预测　；参考：《东北师范大学》2015年硕士论文

【摘要】：为了促进城市交通的可持续发展,必须优先发展公共交通。为候车乘客提供公交车到达时间预报服务有利于提高公共交通服务水平。因此本文利用公交车的GPS数据对公交到达时间预测技术做了研究。由于公交车的行驶受到了很多环境因素的影响,实现公交车到达时间的精确预测是一个复杂而且困难的问题。很多研究者为了解决该难题,提出了很多关于公交车到达时间预测模型,总结当前这些模型主要存在的问题有：未考虑实时路况、模型设计依赖于经验,过拟合,易陷入局部最优,基于一定假设而不能满足实际情况,计算时间复杂度较高而不能满足实时应用需求,模型考虑因素过少,不合理的模型结构使得出现误差累加的情况。基于当前预测模型的研究现状,本文选择支持向量机作为预测模型的理论基础。支持向量机作为一种较新的机器学习算法能完成对复杂的非线性关系的建模,模型设计不依赖于经验和任何假设,不存在过拟合及陷入局部最优的问题。支持向量机在利用已训练好的模型提供预测服务的时间复杂度很低,能满足实时应用需求。由于原始训练集过大导致支持向量机的模型训练不具有可计算性,故本文将对原始训练集进行划分,并以一种多叉树结构组织这些划分后的小训练集,同时这也避免了误差累加的问题。为了提高预测模型的精度,本文考虑了多种影响因素,包括当前时刻的路况信息。本文首先对公交车的运行耗时做了适当的分析,并提出了一种公交线路分段化处理的方案；其次,充分考虑节假日、高峰期、天气、实时路段等因素,提出了一个简单的公交车到达时间预测模型和一个相对复杂的基于E-SVR的公交车到达时间预测模型；最后利用真实的GPS数据并对预测模型的预测精度进行了评估,结果显示提出的预测模型具有较高的预测精度；另外,本文还提出了数据预处理阶段的关键算法：确定公交站点边界、确定车辆进出站时间、确定车辆发车时间。本文以深圳公交为背景,基于本文的研究内容,实现了一个可提供公交车到达时间预报服务的软件系统,并且该系统已经对外提供服务,产生了应用价值。
[Abstract]:In order to promote the sustainable development of urban transportation, priority must be given to the development of public transport. Providing bus arrival time forecast service for waiting passengers is helpful to improve the service level of public transport. In this paper, the GPS data of buses are used to predict the arrival time of buses. Due to the influence of many environmental factors, it is a complicated and difficult problem to realize the accurate prediction of bus arrival time. In order to solve this problem, many researchers put forward a lot of bus arrival time prediction models. The main problems of these models are: not considering the real-time traffic conditions, the model design depends on experience, over-fitting. It is easy to fall into local optimum, based on certain assumptions but can not meet the actual situation, high computational time complexity can not meet the needs of real-time applications, model considerations are too few, unreasonable model structure makes the case of error accumulation. Based on the current situation of prediction model, support vector machine (SVM) is chosen as the theoretical basis of prediction model. As a new machine learning algorithm, support vector machine (SVM) can model complex nonlinear relations. Model design does not depend on experience and any assumptions, and there is no problem of over-fitting and falling into local optimum. Support vector machines (SVM) have low time complexity in providing prediction services using trained models and can meet the requirements of real-time applications. Because the model training of support vector machine is not computable because the original training set is too large, this paper will divide the original training set and organize these small training sets with a multi-tree structure. At the same time, this also avoids the problem of accumulation of errors. In order to improve the accuracy of the prediction model, this paper takes into account a variety of factors, including the traffic information at the current time. In this paper, the running time of the bus is analyzed properly, and a scheme of segmenting the bus line is put forward. Secondly, the factors such as holidays, peak hours, weather and real time sections are taken into full consideration. A simple bus arrival time prediction model and a relatively complex bus arrival time prediction model based on E-SVR are proposed. Finally, the prediction accuracy of the prediction model is evaluated by using the real GPS data. The results show that the proposed prediction model has high prediction accuracy. In addition, the key algorithms in the data preprocessing stage are proposed in this paper: to determine the bus stop boundary, to determine the time of the vehicle entering and leaving station, and to determine the vehicle departure time. In this paper, based on the research content of Shenzhen bus, a software system is implemented to provide bus arrival time forecast service, and the system has already provided the service to the outside, which has produced the application value.
【学位授予单位】：东北师范大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U495

【参考文献】