在智能交通信号控制和交通流诱导系统中,交通环境状态的有效判别是影响交通控制决策的先决条件,本文针对交通流产生的大数据信息,结合深度学习算法提出一种离散化交通状态的判别方法.给出了包括交通状态数据采集、状态数据描述、状态深度学习和判别等功能模块的系统架构,构建了一种离散交通状态编码方法,为深度学习交通状态特征提供了数据基础.模型训练阶段,对采集到的二值和连续值交通状态数据,分别构建了两种不同的深度置信网络实现交通状态特征的无监督学习;模型微调阶段,在整合形成的高层抽象特征向量顶端增加softmax分类器,采用反向传播算法实现参数微调.最后,该方法基于VISSIM微观交通软件进行仿真,实验结果表明,离散交通状态编码方法可有效表达交通状态,基于深度学习的交通状态判别方法相对传统方法具有较高的准确度.%In the intelligent transportation signal control and traffic flow guidance system, effective identification of the traffic state of the environment is a prerequisite for traffic control. Facing large scale of traffic data sets, this paper proposes a identification method of discrete traffic state based on deep learning. The system architecture include four parts of traffic state data acquisition, data description, deep learning and identification. A discrete traffic state encoding method is proposed which provide the base for deep learning of traffic state. In the phase of model training, two deep belief networks including TEDBN and TVDBN are constructed and trained for features learning with the traffic state data of binary value and continuous value. In the phase of model fine-tuning, a softmax classifier is added on the top of all high-level abstraction features, and the back-propagation algorithm is used to fine tune the parameters. At last, the method is validated on the VISSIM simulation software. The experimental results show that the method of discrete traffic state encoding can effectively express the traffic state, and the identification method based on deep learning has higher accuracy than the traditional method on the traffic state.
展开▼
