【作者】 赵玲；

【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】 交通安全是国民经济发展和社会安定的重要方面，也是道路交通管理的两项基本任务之一。道路交通事故预测是道路交通安全研究的一项重要内容，它的目的是为了掌握交通事故的未来状况，以便及时采取相应的对策，有效地控制各影响因素，避免工作中的盲目性和被动性，减少交通事故。道路交通系统的非线性、随机性、动态性以及不确定性等特点，决定了作为道路交通系统行为特征量的道路交通事故预测的复杂性。本文以交通统计年鉴中的资料为数据，以马尔可夫链理论和灰色理论为研究手段，以提高交通事故预测精度和可靠性为目标，详细地探索和研究适应不同道路交通事故特点的交通事故宏观预测方法。论文的主要工作及结论如下：1.针对一般灰色马尔可夫链模型运用的转移概率矩阵固定不变而影响预测精度的问题，通过采用滑动转移概率矩阵方法，建立改进的灰色马尔可夫链模型。借助改进的灰色马尔可夫链模型对全国2002~2004年交通事故10万人口死亡率进行了预测分析。结果表明，改进的灰色马尔可夫链模型的预测精度高于一般灰色马尔可夫链模型的预测精度，具有较强的工程实用性。2.基于交通事故指标为相依随机变量的特点，应用有序聚类的方法划分出反映交通事故衡量指标的变化区间，然后以指标序列规范化后的各阶自相关系数为权重，运用加权马尔可夫模型来预测未来交通事故的状态。以北京市1970~2010年共41年的事故受伤人数为例对该方法进行了预测分析。这种预测方法使得交通事故的预测由点值扩大到区间，大大提高了预测的可靠性。3.鉴于GM(1,1)模型在分析多关联不确定因子系统失效的情况，采用理论基础更扎实的SCGM(1,1)c模型替代GM(1,1)模型，同时结合加权马尔可夫链的优点，构建灰色加权马尔可夫SCGM(1,1)c模型来预测未来时刻的交通事故。以北京市1975~2010年道路交通事故次数为例进行了预测分析。预测结果表明此模型是可靠可信的，具有很强的工程实用性。4.鉴于传统GM(1,1)模型所具有的固有偏差及自身缺陷，采用无偏GM(1,1)模型代替传统GM(1,1),在无偏GM(1,1)模型拟合得到的系统总体趋势的基础上进行马尔可夫预测，结合新信息优先的思想，建立等维新息无偏灰色马尔可夫模型来预测交通事故。以全国2001~2010年道路交通事故死亡人数为例进行了预测分析。预测结果表明它不但短期预测准确度高，而且适合中长期预测。5.针对单一灰色预测模型的假设条件及适用范围均受限的缺点，建立一种基于最优加权组合的预测模型。在权重系数之和为1的约束条件和拟合误差平方和为最小的目标函数下，运用最小二乘原理来求解权重系数。运用最优加权组合模型对我国2001~2010年道路交通事故死亡人数进行了预测分析。结果表明，组合预测模型比单一预测模型能有效降低预测误差，提高预测精度。更多还原

【Abstract】 Traffic safety is an important aspect of national economic development and socialstability, and is also one of the two basic tasks in road traffic management. Road accidentprediction is an important element of road safety research, and its purpose is to grasp thefuture state of traffic accidents, to take corresponding countermeasures timely, to control thevarious influencing factors effectively, to avoid the blindness and passivity, and to reducetraffic accidents. The characteristics such as nonlinear, randomness, dynamic and uncertaintyof road traffic system, determine the complexity of road traffic accidents forecast as one of thebehavioral feature quantities of road traffic system. By means of the data in the transportationstatistical yearbook, and Markov theory and grey theory as research tools, in order to improvethe prediction accuracy and reliability as the goal,this paper discuss and study in detail themacro forecast methods of traffic accidents adapted to the different road characteristics.Themain work and conclusions are as follows:1. Against the problem about prediction accuracy is affected during the generalgrey-Markov chain model for using fixed transition probability matrix; the paper improvesgrey-Markov chain model by using sliding transition probability matrix. By means of theimproved grey-Markov chain model, the100,000population death rate of traffic accidentfrom2002to2004has been predicted. The result shows that the prediction accuracy ofimproved gray-Markov chain model is better than that of the ordinary grey-Markov chainmodel, and the new model has a strong practicability.2. Based on the special characteristics of the annually traffic accidents measurableindicators being a dependent stochastic variables, applying sequential cluster method to set upthe classification standard of traffic accident, regarding the standardized self-coefficients asweights,the paper applies the weighted Markov chain to forecast future state of trafficaccident. This method is used and analyzed during the injuried persons from1970to2010inBeijing.This prediction method makes accident prediction results from the point value to theinterval value, greatly improving the prediction reliability.3. In view of the GM(1,1) model fails to analyze many uncertain factors relatedsystems,the paper uses the SCGM(1,1)c model with solid theoretical basis to replace the GM(1,1). Combining with the advantages of weighted Markov Chain, the grey weightedMarkov SCGM(1,1)c model is built to predict future traffic accident. This method is usedand analyzed during road traffic accidents from1975to2010in Beijing. The results showthat the grey weighted Markov SCGM(1,1)c model is reliable and credible,having a strongpracticability.4. Given the traditional GM(1,1) model has inherent bias and own shortcomings, thepaper substitutes the traditional GM(1,1) model for unbiased grey model. The unbiased greymodel is used to fit the development tendency of the forecast system, while Markovprediction is used to forecast the fluctuation along the tendency. Combined with the idea ofnew information priority, the equal dimension and new information unbiased grey Markovmodel is constructed. This method is applied to predict the deaths from2011to2015with thenumber of road traffic deaths from2000to2010. Experiment results show that the equaldimensional and new information grey Markov forecasting model not only can remainadvantages of short-term forecasting accuracy, but also can improve the medium andlong-term forecast accuracy.5. Considering the single gray prediction model has its own assumptions and limitedscopes, this paper puts forward the combined forecasting model based on optimal weightedmethod. The weight coefficients of combined forecasting model were determined by theprinciple of least squares under the constraint condition that weight coefficients sum is oneand objective function that the fitting error squares sum is minimized. The weightedcombination model was used to predict and analyze the road traffic deaths in the year from2001to2010. The results show that combination forecasting model can effectively reduce theprediction error and improve the prediction accuracy than the single forecasting models.更多还原