【作者】 钱超；

【导师】 许宏科；

【作者基本信息】 长安大学 ， 交通信息工程及控制， 2013， 博士

【摘要】 近年来，随着经济社会的高速发展，我国的高速公路建设事业取得了举世瞩目的成就，但与此同时高速公路也面临着交通需求快速增长与服务能力相对滞后的难题。如何提高高速公路通行效率，有效缓解直至解决收费站区拥堵问题，成为高速公路运营管理领域面临的重要课题。电子不停车收费（ETC）的应用和推广，是解决这一问题的重要举措。随着国内ETC系统的推广，高速公路管理部门积累了大量原始收费数据，在这些记载着车辆通行详细信息的数据内部蕴含着丰富的内在关系和隐含信息，如何从海量数据中获取有效的信息，提高管理决策水平，是亟待解决的关键技术问题。本文研究如何通过对ETC收费数据进行有效整合，利用数据挖掘技术，提取和表达蕴含在原始数据内部的通行行为模式和交通量时空变化趋势，通过对车辆通行行为的预测和异常检测以及交通量预测分析，为高速公路管理部门提供理论依据和决策参考。围绕该研究问题，本文做了以下几方面工作：ETC车辆路径预测与异常检测研究，研究利用ETC收费历史数据，找出车辆通行规律，准确预测车辆的后续通行路径，并从中检测出异常的通行路径；ETC交通量多维预测研究，研究以ETC收费数据为基础，实现多维统计分析，为多维交通量构建预测模型，提出一种基于OLAM实现高速公路交通量多维预测的方法；ETC交通量组合预测模型研究，在单项预测模型基础上设计出一种ETC交通量组合预测模型，充分发挥各单项预测模型的优势，进一步提高预测的精确度和可靠度。本文取得了如下研究成果：1.提出了一种基于混合Markov模型的高速公路车辆路径预测与异常路径检测方法。论文建立了Markov通行行为模型，针对基本Markov模型在预测结果的准确性、覆盖率等方面存在诸多不足之处，提出一种新的Markov模型-混合Markov路径预测模型，给出了使用EM迭代聚类算法对ETC车辆路径序列进行分类的方法，使得同一类车辆具有相同或相似的通行行为，为每类车辆建立独立的Markov模型，用以描述该类别车辆的通行特征，根据历史通行数据预测其后续通行路径。2.构建了一种基于联机分析挖掘（OLAM）利用ETC收费数据实现交通量多维统计分析的模型。论文选取时间、空间、车型、车种等维度为ETC收费数据构建了雪花模型，实现了多维交通量快速汇总统计。3.建立了一种经过异常值修正的季节ARIMA预测模型。论文选取多维统计结果作为序列数据样本，通过对数据样本分别进行平稳化、模型识别、异常值检验、参数估计、模型诊断等步骤，建立经异常值修正的最优季节ARIMA(p, d, q)(P, D, Q)s模型，该模型在预测准确率上优于原始模型，利用该修正模型实现了交通量预测。4.设计了一种基于季节ARIMA模型、BP神经网络和支持向量回归机（SVR）的最优线性组合预测模型。论文以ETC月度交通量为训练样本，分别建立起上述三种单项预测模型；利用单项预测模型预测结果，以预测误差平方和最小为目标函数，建立起求解组合预测权系数的优化模型，根据权系数最优解计算结果，实现组合模型下的月度交通量预测；最后通过建立评价指标体系验证组合模型预测效果优于单项预测模型。更多还原

【Abstract】 In recent years, with rapid development of economic society, expressway constructionin China has made remarkable achievements. As an important part of road transportationsystem，expressway is facing with contradiction between the rapid growth of traffic demandand the relatively lagging of service capability. It has become an important issue forexpressway management operations to improve traffic efficiency and alleviate traffic jameffectively, as well as solve congestion problems in toll station region. Application andgeneralization of electronic toll collection (ETC) system is an important and effectivemeasure to solve this problem. With generalization of ETC system throughout the country,expressway administrators have accumulated large amounts of raw tolling data, these detailtraffic data contain plenty of internal relation and implicit information. How to obtain validinformation from the massive data and improve management and decision level are the keytechnical problems to be solved.This paper focuses on how to effectively integrate raw ETC tolling data, utilizes datamining techniques to extract and express traffic behavior patterns and spatial-temporaltrends of traffic volume, conducts driving behavior prediction and unusual detection, andimplements traffic volume prediction, thus provides theoretical foundation and decisionsupport for expressway operations. Surrounding the problem, several aspects work havebeen studied: ETC vehicle route prediction and unusual detection, which identifies vehicledriving pattern, accurately predicts future driving route and detects abnormal route based onhistorical ETC data; Multi-dimensional prediction of ETC traffic volume, which achievesmultidimensional statistical analysis, formulates prediction model for multi-dimensionaltraffic volume, proposes a method for multi-dimensional prediction of expressway trafficvolume based on Online Analytical Mining (OLAM); Combinational ETC traffic volumeprediction model, which proposes an combination model based on single prediction models,takes full advantages of each single model to further improve accuracy and reliability ofprediction. The main research results are as follows: 1. method of vehicle route prediction and abnormal path detection are proposed inthis paper based on hybrid Markov model. Against the shortcomings of low accuracy andcoverage rate of basic Markov route prediction model, this paper introduces a new Markovmodel (a hybrid Markov route prediction model) and provides a method for classifying ETCvehicle route sequences using EM iterative clustering algorithm so that vehicles in the sameclass have the same or similar driving behavior; it also builds independent model for eachclass of vehicles to describe its driving behavior. Finally, the paper predicts the futuredriving route using historical data.2. This paper constructs a model using ETC tolling data to implementmulti-dimensional prediction of expressway traffic volume based on OLAM. Time, spaceand other dimension information are selected to formulate snowflake schema of ETC dataand get multidimensional statistics of traffic volume.3. This paper builds seasonal ARIMA(p, d, q)(P, D, Q)s model with its abnormalvalue corrected. First we selects multidimensional statistical results as sequence data sample,then conducts smoothing, model identification, outlier test, parameter estimation, modeldiagnostics and other steps on data samples respectively, thus establishes optimal seasonalARIMA(p, d, q)(P, D, Q)s model with its abnormal value corrected. Finally,multidimensional prediction of traffic volume is realized by using this predicting model, andresults show that its prediction accuracy is better than original model.4. On the basis of seasonal ARIMA model, BP neural network and support vectorregression (SVR), an optimal linear combination prediction model is proposed. This paperemploys ETC monthly traffic volume as training samples to build these three singleprediction model; then takes minimizing the sum of squared errors as objective function anduses results of single prediction model to build optimal model of calculating combinationweight coefficients, realizes the prediction of monthly traffic volume according to results ofweight coefficients calculation; Finally verifies that combination prediction model is betterthan signal prediction model by establishing evaluation index system.更多还原