基于多维数据的行程时间预测与可靠性研究

发布时间：2018-07-13 21:17

【摘要】：随着交通事业的发展和路网的延伸,高速公路发展水平与公众交通需求呈现双增长趋势,交通信息服务日趋个性化、差别化、精细化。如何对不同车型用户,不同气象场景,不同时段等因素影响下的多维交通信息进行挖掘与信息发布,成为新的研究课题。现代快节奏的生活中,时间价值日益受到重视,出行者越来越关注行程时间的延误和可靠性。及时发布行程时间预测值及其可靠度,能为驾驶者的路径选择提供支持。基于此,本文利用收费数据与气象监测数据,以辽宁省高速公路为试验路段,开展行程时间预测及其可靠性研究,为公众出行及相关部门管理运营提供决策依据。主要完成的工作及取得的成果包括:(1)研究分析车型、时间、气象因素对行程时间的影响,设计了以行程时间为主题的多维数据仓库逻辑模型,搭建数据仓库结构框架。针对非同源数据的集成问题,提出了一种时空匹配法,实现联网收费数据与气象监测数据的集成。提出数据清洗、数据转换的方法,清除了异常数据,实现了数据格式统一,完善了收费数据仓库。(2)研究行程时间稀疏数据和异常数据的处理方法。创新性地提出了"上、下游数据构造法",解决了收费数据的稀疏问题。在相关研究的基础上提出了"改进四分法"的数据筛选方法,有效剔除了数据中的离群值。处理后,数据信息更加完整,贴近真实情况。设计了行程时间序列的提取方法,为行程时间预测研究做准备。采用OLAP联机分析处理技术,提取多维行程时间信息,定量地分析了时间、车型、气象等因素对行程时间的影响,验证了分维度研究的合理性。(3)研究行程时间序列的自相关与偏自相关特性,利用BIC准则实现模型定阶,利用最小二乘法进行参数估计,建立ARMA行程时间预测模型。增设车流量序列作为回归变量,建立ARMAX预测模型,改善了传统ARMA模型的预测效果。案例表明,ARMAX的行程时间预测效果良好,能够满足实际需要。改善了传统ARMA模型的预测滞后问题,最大百分比误差较传统ARMA模型的降低约5%。(4)研究历史行程时间的分布特征,利用多种概率模型进行拟合,经过K-S假设检验与拟合优度的对比,证明对数正态分布是行程时间可靠性的最优表达模型。基于该模型,确立了行程时间可靠性测度指标的计算方法。选择变异系数、缓冲指数、计划时间指数、拥挤频率作表征历史行程时间可靠度的指标,运用实例研究了车型、时间、气象等多维因素对可靠度的影响。提出了预测行程时间可靠概率及预测行程时间缓冲指数,所提出的指标将未来行程时间与历史统计行程时间相结合,实现对未来行程时间可靠程度的评价,补充了行程时间可靠性指标体系。案例表明预测行程时间可靠性指标对指导路径决策,引导公众出行具有实际意义和重要作用。
[Abstract]:With the development of traffic and the extension of road network, the development level of expressway and the demand of public transportation are increasing, and the traffic information service is becoming individualized, differentiated and refined. How to mine and publish multi-dimensional traffic information under the influence of different vehicle users, different weather scenes and different time periods has become a new research topic. In modern fast-paced life, the value of time is paid more and more attention, and travelers pay more and more attention to the delay and reliability of travel time. Timely release of travel time prediction value and its reliability can provide support for driver's path selection. Based on this, this paper makes use of toll data and meteorological monitoring data, taking Liaoning Expressway as the experimental section, carries out travel time prediction and reliability research, and provides the decision basis for public travel and the management and operation of relevant departments. The main work and achievements are as follows: (1) the effects of vehicle, time and meteorological factors on travel time are analyzed. A multi-dimensional data warehouse logical model with travel time as the theme is designed, and the data warehouse structure framework is built. To solve the problem of integration of non-homologous data, a spatio-temporal matching method is proposed to realize the integration of network toll data and meteorological monitoring data. The methods of data cleaning and data conversion are put forward, the abnormal data is eliminated, the data format is unified, and the data warehouse is improved. (2) the processing methods of travel time sparse data and abnormal data are studied. The upstream and downstream data construction method is innovatively proposed, which solves the sparse problem of charge data. On the basis of related research, an improved quadrilateral data screening method is proposed, which can effectively eliminate outliers in the data. After processing, the data information is more complete, close to the real situation. The extraction method of travel time series is designed to prepare for the study of travel time prediction. OLAP OLAP OLAP technology is used to extract multidimensional travel time information and quantitatively analyze the influence of time, vehicle type, weather and other factors on travel time. (3) the autocorrelation and partial autocorrelation characteristics of travel time series are studied. BIC criterion is used to determine the order of the model, the least square method is used to estimate the parameters, and the ARMA travel time prediction model is established. The ARMAX prediction model is established by adding the traffic flow sequence as a regression variable, which improves the prediction effect of the traditional ARMA model. The case shows that the travel time prediction of ARMAX is effective and can meet the actual needs. The prediction lag problem of the traditional ARMA model is improved, and the maximum percentage error is reduced by about 5 times compared with the traditional ARMA model. (4) the distribution characteristics of the historical travel time are studied and fitted by various probability models, and the K-S hypothesis test is compared with the goodness of fit. It is proved that the lognormal distribution is the optimal representation model of travel time reliability. Based on this model, the calculation method of travel time reliability measure index is established. The variation coefficient, buffer index, planning time index and congestion frequency are selected as indicators to characterize the reliability of historical travel time. The effects of multi-dimensional factors such as vehicle type, time and meteorology on reliability are studied with examples. The reliability probability of predicting travel time and the buffer index of predicted travel time are proposed. The proposed index combines the future travel time with the historical statistic travel time to realize the evaluation of the reliability of the future travel time. The reliability index system of travel time is supplemented. The case shows that the reliability index of predicting travel time has practical significance and important role in guiding path decision and guiding public travel.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U491

【参考文献】