基于SVR的数据预处理分析与研究

发布时间：2018-01-11 21:23

本文关键词：基于SVR的数据预处理分析与研究　出处：《兰州理工大学》2014年硕士论文　论文类型：学位论文

【摘要】：随着我国经济的增长和城市化进展,交通拥堵、交通事故频发、尾气污染等交通问题已经成为当今社会普遍关注的焦点。实时而准确的短时交通流量预测可以为城市交通诱导和控制提供数据支持,是解决多种交通问题的关键和基础。本论文针对断面交通流预测数据中往往存在的错误、缺失、包含较多噪声等问题,结合SVR预测模型,提出了一种新的数据预处理方法,根据路网交通流信息中隐含的时空关系,增加了对目标路段上游交通流数据在时间、空间上的相关性分析和处理,其优点在于降低了预测过程中的不确定性,适应了交通流的随机变化,并结合支持向量回归机所具备的推广能力和对小样本数据具有的较强适应性,提高了预测的精度与泛化能力。最后本论文结合常用的数据预处理技术,对比未使用本论文预处理的SVR模型和神经网络模型,验证了使用本论文方法的模型拟合度有明显的提高,均方误差也明显减小,并且得出了最优的预测方案。通过对目标路段上游的交通流数据进行时空相关性分析,选取对目标路段交通流预测影响较大的路段和其相关数据,并基于线性回归的思想构建模型,将该模型计算的结果运用到SVR模型的数据集中,在避免了数据丢失的同时,既有效的压缩了数据集特征数,降低了计算量,也提高了在预测模型的预测精度和泛化能力。实验结合ε-SVR模型,验证了本论文预处理方法的有效性,并且显著提高了原模型的预测精度,减少了预处理模型的待估参数,提高了模型的计算效率。
[Abstract]:With the economic growth and urbanization in China, traffic congestion and traffic accidents occur frequently. Emission pollution and other traffic problems have become the focus of attention in the society. Real-time and accurate short-term traffic flow prediction can provide data support for urban traffic guidance and control. It is the key and foundation to solve a variety of traffic problems. This paper combines the SVR forecasting model to solve the problems of error, lack and noise in the cross-section traffic flow prediction data. A new data preprocessing method is proposed. According to the spatial and temporal relationship implied in the traffic flow information of the road network, the correlation analysis and processing of the upstream traffic flow data of the target section in time and space are added. It has the advantages of reducing the uncertainty in the prediction process, adapting to the random change of traffic flow, and combining the generalization ability of support vector regression machine and the strong adaptability to small sample data. The prediction accuracy and generalization ability are improved. Finally, the SVR model and the neural network model which are not preprocessed in this paper are compared with common data preprocessing techniques in this paper. It is verified that the fitting degree of the model using the method in this paper is obviously improved and the mean square error is obviously reduced, and the optimal prediction scheme is obtained. Through the spatio-temporal correlation analysis of the upstream traffic flow data of the target section, the section and its related data which have a great influence on the traffic flow prediction of the target road section are selected, and the model is constructed based on the idea of linear regression. The result of this model is applied to the data set of SVR model, which can not only avoid the data loss, but also effectively compress the feature number of the data set and reduce the calculation amount. It also improves the prediction accuracy and generalization ability of the prediction model. Experiments combined with 蔚 -SVR model verify the effectiveness of the preprocessing method and improve the prediction accuracy of the original model significantly. The estimated parameters of the preprocessing model are reduced, and the calculation efficiency of the model is improved.
【学位授予单位】：兰州理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U495;TP18

【参考文献】