摘要
事件检测算法是交通事件管理系统的关键技术之一,提出一种基于树增强朴素贝叶斯(TAN)分类算法对交通事件进行检测,它的网络结构和参数通过数据学习确定,相比贝叶斯网络算法,对专家经验依赖较小。采用小波去噪、标准化和基于熵的离散化方法对原始交通数据进行预处理,将交通事件作为"0-1"分类变量,交通特征参数作为属性变量,构建TAN分类器。采用新加坡艾耶尔国王高速公路(AYE)的数据集对该算法进行了实例验证,实验结果表明TAN分类算法与多层前馈神经网络(MLF)算法的检测性能相当,它们的检测率分别为95.97%和98.8%,但TAN分类算法在模型训练和标定的速度上具有显著优势,且相比MLF算法,TAN分类算法的原理更加简单易懂,因此TAN分类算法具有更广泛的应用前景。
Algorithms for traffic incident detection is one of core techniques in traffic incident manage systems.An algorithm based on tree augmented naive Bayesian(TAN)classifier is developed for traffic incident detection.The network structure and parameters of TAN classifier are both from data learning.Compared with the algorithms based on Bayesian Network,this algorithm has less dependency on knowledge of experts.The raw traffic data is pre-processed by methods of wavelet denoising,normalization,and entropy based discretization.To develop the TAN classifier,traffic incident is regarded as a“0-1”classification variable,and traffic characteristic parameters are regarded as attribute variables.A simulation dataset on the section of the Ayer Rajah Expressway(AYE)in Singapore is taken as a case study to verify this algorithm.The results show that the algorithm based on TAN classifier has a similar performance to the algorithm based on multi-layer feed forward(MLF)neural networks,their detection rates are 95.95%and 98.8,respectively.However,the algorithm based on TAN classifier has a significant superiority on the speed of model training and calibration,and the theory of it is much less complicated than which of MLF.The algorithm based on TAN classifier has a wider application.
作者
凃强
李大韦
程琳
TU Qiang;LI Dawei;CHENG Lin(School of Transportation,Southeast University,Nanjing 210096,China)
出处
《交通信息与安全》
CSCD
北大核心
2018年第3期27-32,共6页
Journal of Transport Information and Safety
基金
国家自然科学基金青年科学基金项目(51608115)
国家自然科学基金国际合作与交流项目(51561135003)
江苏省基础研究计划(自然科学基金)项目(BK20150613)资助
关键词
交通管理
事件检测
贝叶斯网络
熵
TAN分类器
traffic management
incident detection
Bayesian Network
entropy
TAN classifier
