集成随机森林的交通拥堵检测模型

Traffic Congestion Detection Model Based on Ensemble Random Forests

根据现有的城市交通网拥堵检测体系,针对现有方法处理交通网格监测数据流难以获得相对稳定的准确率的问题,提出了一种集成随机森林的交通拥堵检测模型;该模型通过将多个随机森林分类器进行集成实现了交通网分布式监测数据流的并行处理,设计了二级级联分类器对交通网状态进行判定,并可对各监控节点权重进行评估;模型实现主要分为特征提取、集成建模和结合分析3个步骤;在不同规模的交通状态监测网络下分析了模型的综合性能,并分别与其它主流方法进行了对比;实验表明:提出模型具有更好的交通网监测数据流的处理能力,且具备较好的扩展和裁剪性能;该模型提供了一种可应用的交通拥堵检测方法。

According to the existing traffic congestion detection system of cities, a detection system is proposed to solve the problems of relatively low and unstable accuracy in processing the traffic monitoring data. This model integrated multiple random forests （RF） to process each node data in the traffic network parallel, then a cascade classifier is designed to recognize the traffic network status. At last, the importance of node in the traffic network is assessed by using RF. The implementation of this model mainly consisted of three levels, that is, fea ture extraction, building the integrated classification model and combination analysis. Comprehensive performance of the model is analyzed under different size traffic network, and compared respectively with other algorithms. Finally, experiments show the proposed model not only has better comprehensive performance in traffic network monitoring data, but also can be adapt to the change of network size. This model provides an application model for traffic congestion detection.