城市交叉口短时交通流预测模型与算法研究

发布时间：2018-07-03 03:40

本文选题：短时交通流预测 + 时空依赖性　；参考：《兰州交通大学》2014年硕士论文

【摘要】：城市交通问题早已升级为城市可持续发展的最大制约。智能交通系统ITS恰是能够解决这一问题的对症方法，，实时准确的流量预测信息是实现有关于ITS中动态路径诱导系统的基础和关键，而交叉口是道路网中道路通行能力的咽喉、交通阻塞和事故的多发地，因此对交叉口的交通流量的预测显得越发重要。目前交通流诱导控制的时间跨度变短，使得交通流量变化的随机性、混沌性、非线性和不确定性加强，导致早期的有检测器交叉口的短时交通流量预测模型开始不能够很好的契合交通流量变化时的种种特性，也不能够较好的规避随机因素对流量的预测产生较大的影响，得到的结果不能令人满意；而在大多数城市有检测器的交叉口数量还达不到全部交叉口的十分之一，如此一来对于没有安装检测器的交叉口的交通流信息就很难取得，以上这些均成为了城市交通系统早日实现智能化的障碍与瓶颈，因此，针对两类交叉口实时准确的短时交通流量预测的研究显得尤为急切与重要。在本文中首先分析和研究了国内外学者针对两类交叉口短时交通流量预测的现状、未来的发展趋势、存在的问题。共归结了三大问题以便在文章进行研究解决。其次，利用递归图和Lyapunov指数对费家营交叉口东进口的交通流数据进行了可预测性和混沌性分析。在此基础上，再对该交叉口短时交通流的时空依赖性进行研究。第一，在时间维度上利用相似及波动系数进行了交通流的周相似性研究，确定了工作日、休息日、天气（如晴天、雨天）等时间因素在交通流预测中的重要作用。第二，在空间维度上确定了交叉口进出口处预测断面与周边交叉口及路段之间流量的相互影响关系，并针对邻接和非邻接路段将其空间依赖性量化，得到空间上的影响波及范围。通过以上研究提出了基于多维时空参数的短时交通流预测模型及框架，为后面交叉口短时交通流量预测提供依据。然后，针对有检测器交叉口交通流量预测从组合模型的搭配模式和单项模型的权重参数选取方面着手对模型进行改进和优化。依照各单项模型的优缺点，选取三大子模块并将其进行改造以便能够利用多维时空因素；由于预测误差为随机误差，则利用正态分布的良好特性，提出了基于反馈机制的德尔塔变权重法，即利用各子模块的预测误差加权平均的方法，对于预测精度较高的预测值赋予较大的权重，由于各时段的交通状态关系误差不断的变化进行反馈，从而权重可及时进行更新调整，不会造成过大的预测偏差，从而建立起了基于时空关联状态组合预测模型，并以安宁区区域路网部分交叉口为例，验证了模型和算法的可行性。最后，针对无检测器交叉口交通流的预测问题，通过建立无检测器交叉口与有检测器交叉口之间的联系，从而利用有检测器交叉口流量来进行预测的角度出发。第一，介绍了常用的几类归类方法，引入了基于贝叶斯最小风险准则的PNN概率神经网络这一概念，并首次将其应用在无检测器交叉口与有检测器交叉口的归类中，提出了基于PNN概率神经网络的交叉口分类模式的预测方法。第二，介绍了归类后的预测手段，即有线性回归和非线性拟合，并引出了BP神经网络和遗传GA优化后的BP神经网络的两类非线性拟合思路。建立了有检测器交叉口和无检测器交叉口动态联系，实现无检测器交叉口在时间空间上的归类与短时交通流量数据的预测，并以安宁区区域路网部分交叉口为例，验证了模型和算法的可行性。
[Abstract]:The urban traffic problem has been upgraded to the biggest restriction of urban sustainable development. The intelligent transportation system ITS is the right way to solve this problem. Real-time accurate traffic prediction information is the basis and key to realize the dynamic path induction system in ITS, and the intersection is the throat of road traffic capacity in the road network. It is more important to predict the traffic flow of the intersection, so the time span of the traffic flow induced control is shorter, which makes the traffic flow change randomness, chaos, nonlinearity and uncertainty strengthened, which leads to the inability to predict the short time traffic flow prediction model at the early stage of the detector intersection. It is good enough to fit the characteristics of the change of traffic flow, and it can not be better to avoid the large impact of random factors on the prediction of flow, and the results can not be satisfactory. In most cities, the number of intersections of the detector is not up to 1/10 of the entire intersection, so that there is no installation inspection. The traffic flow information at the intersections of the measuring apparatus is difficult to obtain. These all become the obstacles and bottlenecks of the early realization of the intelligent urban traffic system. Therefore, the research on real-time and accurate short-term traffic flow prediction for the two types of intersections is particularly urgent and important.
In this paper, we first analyze and study the current situation of short-term traffic flow forecast at the two types of intersections at home and abroad, the future development trend and the existing problems. Three major problems are summed up in order to solve the problem in this article.
Secondly, the predictability and chaos of traffic flow data imported from the east of Fisher intersection are analyzed by using recursion and Lyapunov index. On this basis, the temporal and spatial dependence of short time traffic flow at the intersection is studied. Firstly, the cycle similarity of traffic flow is studied by using similarity and wave coefficient in time dimension. The important role of time factors such as working days, rest days, weather (such as sunny days, rainy days) and other factors in traffic flow prediction is determined. Second, the interaction relationship between the prediction section of the inlet and exit of the intersection and the flow between the surrounding intersections and the sections is determined on the spatial dimension, and the spatial dependence of the adjacent and non adjacent sections is quantified. Through the above study, a short-term traffic flow prediction model and framework based on multidimensional space-time parameters is proposed, which provides the basis for the short-term traffic flow prediction in the rear intersection.
Then, the model is improved and optimized for the combination model of the detector intersection and the selection of the weight parameters of the single model. According to the advantages and disadvantages of each single model, three modules are selected and modified to make use of the multidimensional time and space factors. According to the good characteristics of the normal distribution, the Del tower variable weight method based on the feedback mechanism is proposed, that is, the weighted average of the prediction error of each sub module is used to give greater weight to the predicted value with higher prediction accuracy. The renewal and adjustment can be carried out in time, which will not cause excessive prediction deviation, thus the combined prediction model based on spatio-temporal correlation state is established, and the feasibility of the model and algorithm is verified by the example of the part intersection of the regional road network in the Anning area.
Finally, in view of the prediction problem of traffic flow without detector intersection, by establishing the connection between the detector intersection and the intersections of the detector, the prediction angle is made using the traffic of the detector intersection. First, several commonly used classification methods are introduced, and the PNN based on the Bayes minimum risk criterion is introduced. The concept of rate neural network is used for the first time in the classification of intersections without detector intersection and detector. The prediction method of intersection classification based on PNN probabilistic neural network is proposed. Second, the prediction method after classification is introduced, that is, linear regression and nonlinear fitting, and leads to BP neural network and heredity. The two kind of nonlinear fitting method of BP neural network after GA optimization is proposed. The dynamic connection between the detector intersection and the non detector intersection is established to realize the classification of the non detector intersection in time and space and the prediction of the short time traffic data, and the feasibility of the model and the algorithm is verified by the example of the part intersection of the regional road network in the Anning area. Sex.
【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U491.112

【参考文献】