基于时空关联和BP_Adaboost的短时交通参数预测  

Short-term Traffic Parameter Prediction Based on Spatio-temporal Correlation and BP_ Adaboost

作　　者：丁闪闪[1,2] 王维锋[1,2] 季锦章[1,2] 党倩[1,2] 

机构地区：[1]江苏省交通规划设计院股份有限公司,江苏南京210014 [2]交通运输部智能交通技术和设备行业研发中心,江苏南京210014

出　　处：《公路交通科技》2016年第5期98-104,共7页Journal of Highway and Transportation Research and Development

基　　金：国家科技支撑计划项目(2014BAG01B02);江苏省交通科学研究计划项目(2012X07-1);江苏省交通规划设计院博士后基金项目(KY2013022)

摘　　要：为提高预测精度和降低计算复杂度,提出了一种基于时空关联和BP_Adaboost的短时交通参数预测方法。首先,通过分析交通参数之间的时间关联性和空间关联性,得到对预测值产生影响的因素。然后,采用主成分分析法对影响因素进行预处理,选取了主成分作为后续预测模型的输入。最后,采用Adaboost算法对BP神经网络进行优化,构建了BP_Adaboost预测模型,以改善传统BP神经网络的预测性能。利用南京市典型道路数据对该方法进行了性能评价。结果表明:该方法不仅可以在一定程度上提高预测精度,更重要的是能够大大提高预测结果的稳定性。由此可见,该方法可以作为短时交通参数预测的有效手段。A short-term traffic parameter prediction approach based on spatio-temporal correlation and BP_Adaboost is proposed to improve prediction accuracy and reduce the computational complexity. First,time correlation and space correlation among traffic parameters are analyzed to obtain the influencing factors on the predicted values. Then,these influencing factors are preprocessed by using PCA method,and the primary components are chosen as the input of the following prediction model. At last,the BP neural network is optimized with Adaboost algorithm,the BP_ Adaboost prediction model is built to improve the prediction performance of traditional BP neural network,and this approach is evaluated by using the typical data of the roads in Nanjing. The result shows that the approach not only can improve the prediction accuracy in some extent,but more importantly it can greatly improve the stability of the prediction result. Thus,the proposed approach can be used to predict shot-term traffic parameters effectively.

关 键 词：交通工程 交通参数 主成分分析 ADABOOST算法 BP神经网络 短时预测 时空关联性 

分 类 号：U491.14[交通运输工程—交通运输规划与管理]