基于车辆构成和特征提取的交通状态预估模型

Traffic State Estimation Model Based on Vehicle Composition and Feature Extraction

准确的交通状态预估有利于车辆选择合理交通路线,缓解交通拥堵状态。针对传统方法特征提取不充分、预估准确度依赖于监测器精度等问题,提出了一种基于车辆构成和特征提取的交通状态预估模型。该模型以速度、流量、占用率和大型车辆比重为输入,将交通状态分成畅通、拥挤和拥堵三种状态,结合时间空间维度来预估交通状态。通过卷积神经网络(ConvolutionalNeural Network,CNN)提取交通拥堵特征,得到的特征输入支持向量机(Support Vector Machine,SVM)进行交通状态预估。实验表明,考虑车辆构成比忽略车辆构成准确率提高1. 12%, CNN-SVM模型预估准确度比CNN模型提升2. 25%,是一种有效的交通状态预估模型。

Accurate traffic state estimation is conducive to the vehicle to choose a reasonable traffic route and ease traffic congestion. In order to solve the problems that the traditional method extracts traffic congestion feature insufficiently and the estimation accuracy depends on the accuracy of the monitor, a traffic state estimation model based on vehicle composition and feature extraction is proposed. The model used speed, flow, occupancy and large vehicle weight as input, and divided the traffic state into three states: smooth, crowded and congested, and combined the time and space dimensions to estimate the traffic state. The convolutional neural network(CNN) extracted traffic congestion features automatically which would be used in SVM for traffic state estimation. Experiments show that the model considering vehicle composition increases the accuracy by 1. 12% compared with the model that ignores the vehicle composition. The CNN-SVM model estimation accuracy is 2. 25%higher than the CNN model, which is an effective traffic state estimation model.