基于浮动车速度波动特征的交通状态识别

为实现路段交通状态的准确判别，解决单参数无法直接识别道路交通状态问题，本文利用高频浮动车速度数据，使用灰度共生矩阵特征值对比度和逆方差表示车辆行驶的波动特征。基于城市道路交通状态变化的动态性与连续性，围绕固定时间窗口内车辆的平均车速、对比度和逆方差，采用FCM (Fuzzy c-means)算法进行聚类分析，得到畅通、平稳、拥挤和阻塞这4种状态阈值。提出基于多维高斯隐马尔可夫模型的交通状态识别方法，分别以3,5,6 min固定时间窗口训练模型。模型状态转移矩阵表明，时间窗口越小其保持原有交通状态的可能性越大，时间窗口越大交通状态突变的可能性越大。使用不同序列长度对比3种时间窗口在测试集中的识别精度，结果表明，随着序列长度的变化，精度显示出先升高后降低的趋势，且固定时间窗口越大，不同序列长度的识别精度变化越均匀。最后利用5 min固定时间窗口划分数据使用本文方法和支持向量机以及随机森林分别进行道路交通状态识别，综合精度分别为92.00%、84.89%、88.48%，同时本文方法在查准率、召回率和F1度量(F1-score)指标均优于其他两个模型，说明道路车速的波动特征可以很好地反映道路交...

Traffic State Recognition Based on Speed Fluctuation Characteristics of Floating Car

To realize the accurate identification of the road traffic state and solve the problem that the single parameter cannot directly identify the road traffic state, this paper uses the high-frequency floating vehicle speed characteristic data and the gray co- occurrence matrix eigenvalue contrast and inverse variance to represent the fluctuation characteristics of vehicle driving. Based on the dynamic and continuous changes of urban road traffic state, the average speed, contrast, and inverse variance of vehicles in a fixed time window are analyzed using the Fuzzy c-means (FCM) algorithm, and four state thresholds are obtained: free, smooth, crowded and blocked. A traffic state recognition method is proposed based on a multi-dimensional Gaussian Hidden Markov model. The model is trained with fixed-time Windows of 3 min, 5 min, and 6 min respectively. The state transition matrix of the model shows that the smaller the time window is, it is more likely to keep the original traffic state, and the larger the time window, it is more likely to change the traffic state. Using different sequence lengths to compare the recognition accuracy of the three-time Windows in the test set, the results show that the accuracy increases first and then decreases with the change of sequence length, and the larger the fixed time window, the more uniform the change of recognition accuracy of different sequence lengths. At last, the 5 min fixed time window was used to partition the data, and the proposed method, support vector machine, and random forest were used to identify the road traffic state, and the comprehensive accuracy was respectively 92%, 84.89% and 88.48%. By comparing the precision, recall, and F1 measurement of each state, the proposed method is better than other two models, which indicates that the fluctuation characteristics of road speed can well reflect the road traffic state, and the multi-dimensional Gaussian Hidden Markov model has a good effect on the recognition of road traffic state.