考虑大型车因素的支持向量机短时交通状态预测模型研究  

Study on Short-term Traffic State Forecasting Model of SVM Considering Proportion of Large Vehicles

作　　者：孙静怡[1] 牟若瑾 刘拥华[1] SUN Jing-yi;MOU Ruo-jin;LIU Yong-hua(School of Transportation Engineering,Kunming University of Science and Technology,Kunming Yunnan 650500,China)

机构地区：[1]昆明理工大学交通工程学院,云南昆明650500

出　　处：《公路交通科技》2018年第10期126-132,共7页Journal of Highway and Transportation Research and Development

摘　　要：高速公路是运输效率高、通行能力大、具备全控制条件的道路,及时发现异常交通状态并采取相应的管制措施是保持良好通行能力的基础。交通状态的短时预测技术能够在实时交通状态数据的基础上对交通状态发展趋势进行预测,为高速公路主线运行管理及匝道交通控制提供决策依据。本研究首先应用灰色关联度理论,分析了道路车型组成比例对交通状态的影响,发现车流中的大型车比例与车流平均速度存在较强的关联性,而且在纵断面线形较为复杂的山区高速公路,车流中大型车比例对车流速度的影响更为显著;然后建立了引入大型车辆比例影响因素的基于支持向量机(SVM)模型的高速公路短时交通状态预测模型,最后通过实测数据及对比试验验证了模型的可行性与准确性。结果表明:本研究设计的支持向量机预测模型具有较为准确的预测效果,均方误差为0. 024 19,决定系数为0. 58;与未引入大型车辆比例的预测方案相比,均方误差减少0. 22,决定系数增大0. 27;与传统的BP神经网络模型相比,支持向量机短时交通状态预测模型预测结果震荡幅度小,所需训练样本量少,具有良好的预测精度,综合性能较好;通过时间序列分析得到,以前6,7个时间点作为输入的预测方案效果较为准确,若时间选取过多,将对模型产生干扰,预测效果反而不好。Expressway is a kind of road with high efficiency,large capacity and full control,it is the basis to keep high capacity through finding abnormal traffic status and taking corresponding control measures timely.Short-term traffic state forecasting techniques can forecast the development trend of traffic status based on the real-time traffic status data,and it can provide support for decision-making of management and control of the main lane and on-ramp operation. By applying the grey correlation theory,the influence of proportion of road vehicles on traffic state is analyzed,it is found that there is a strong correlation between the proportion of large vehicles in flow and the average speed of traffic. The influence is more significant on traffic speed in the mountainous area expressway that have complex profile alignment. Then,the expressway short-term traffic state forecasting model based on SVM with the influencing factor of the proportion of large vehicles is built,and the feasibility and accuracy of the model are verified by measured data and contrast experiment. The result shows that（ 1） the designed SVM forecasting model has accurate prediction effect,the mean square error is 0. 024 19 with determination coefficient of 0. 58;（ 2） compared with the prediction scheme withoutintroducing the proportion of large vehicles,the mean square error reduced by 0. 22,and the determination coefficient increased by 0. 27;（ 3） compared with the traditional BP neural network model,the SVM model for short-term traffic state prediction model has low forecasting oscillation amplitude,less training samples,and good forecasting accuracy and comprehensive performance;（ 4） through time series analysis,it is found that the prediction scheme with the previous 6 or 7 time points as input has a relatively accurate effect,while it will interfere the model and the prediction effect is not good if too much time is selected.

关 键 词：交通工程 短时交通状态预测 支持向量机 高速公路 灰色关联度 

分 类 号：U491.14[交通运输工程—交通运输规划与管理]