交通运输系统工程与信息 ›› 2015, Vol. 15 ›› Issue (4): 51-56.

• 智能交通系统与信息技术 • 上一篇    下一篇

高斯过程回归短时交通流预测方法

康军*1,2,段宗涛1,2,唐蕾1,刘研1,王超1   

  1. 1. 长安大学信息工程学院,西安710064;2. 陕西省道路交通智能检测与装备工程研究中心,西安710064
  • 收稿日期:2015-03-25 修回日期:2015-05-30 出版日期:2015-08-25 发布日期:2015-08-25
  • 作者简介:康军(1975-),男,陕西咸阳人,副教授,博士.
  • 基金资助:

    国家自然科学基金(61303041);交通运输部基础研究项目(2014319812150);陕西省工业科技攻关项目(2014K05-28,2015GY002);中央高校科研资金项目(2013G2241020,2013G1241119);中央高校创新团队项目(310824153405)

A Short Term Traffic Flow Prediction Method Based on Gaussian Processes Regression

KANG Jun1,2,DUAN Zong-tao1,2,TANG Lei1,LIU Yan1,WANG Chao1   

  1. 1. The Information Technology School, Chang'an University, Xi’an 710064, China; 2. Shaanxi Road Traffic Detection and Equipment Engineering Research Center, Xi'an 710064, China
  • Received:2015-03-25 Revised:2015-05-30 Online:2015-08-25 Published:2015-08-25

摘要:

已有的短时交通流预测方法均属于确定性预测,无法对预测的不确定性进行定量分析.针对上述问题,提出了一种基于高斯过程回归的短时交通流预测方法.通过该方法在对短时交通流进行预测的同时还可以得到预测的方差估计值,并依此可以确定预测值的95%置信区间.在仿真实例中,在相同条件下对所提方法与支持向量机预测方法进行比较.仿真结果表明,高斯过程回归短时交通流预测方法不仅与支持向量机预测方法具有相近的预测精度,其中均方根误差为12.09,绝对值误差为118.42,相对误差为17.32%,而且能够获得预测结果的方差估计值,从而有效实现短时交通流概率意义上的预测.

关键词: 智能交通, 短时交通流预测, 高斯过程回归, 短时交通流, 概率性预测, 方差估计

Abstract:

The previous methods about short term traffic flow prediction have been all belong to the deterministic prediction, and cannot be used for the quantitative analysis of the uncertainty of the prediction. Aiming at above problem, a short term traffic flow prediction method based on Gaussian processes regression is presented. Using the proposed method, the short term traffic flow predicting and estimate of variance can be obtained simultaneously, and the 95% confidence interval about this prediction is further calculated. Under the same condition, two simulation tests are realized for the proposed method and the support vector machine prediction method respectively. The test results indicate that the prediction accuracy of the proposed method is similar to the support vector machine prediction method, the root mean square error is about 12.09, the absolute value of error is about 118.42, and the relative error is about 17.32%. Furthermore, using the proposed method, the estimate of prediction variance can be achieved, by which the probabilistic prediction of the short-term traffic flow is implemented effectively.

Key words: intelligent transportation, short term traffic flow prediction, Gaussian processes regression, short term traffic flow, probabilistic prediction, variance evaluation

中图分类号: