基于时空特性和组合深度学习的交通流参数估计

为深入挖掘交通流时空特性，提高交通流参数估计精度，基于深度学习提出一种交通流参数估计的组合方法。根据目标断面及其上游断面的交通流数据构造输入矩阵，利用卷积神经网络捕捉交通流的空间特性，使用长短期记忆和门控循环神经网络挖掘交通流的时间特性，组合3 种深度学习方法所得输出，得到交通流参数估计值。采用中国安徽省合肥市和美国加州萨克拉门托的交通流数据进行验证。结果表明：新方法的性能优于已有各种方法，使估计误差降低 5.72%~33.29%；新组合方法具有较高的准确性和可靠性，能为智能交通系统运营与管理提供高质量的基础数据。

Traffic Flow Parameters Estimation Based on Spatio-temporal Characteristics and Hybrid Deep Learning

To explore the spatio-temporal characteristics of traffic flow and improve the estimation precision, this paper proposes a hybrid deep learning method for traffic flow parameters estimation. The input matrix was constructed by the traffic flow data obtained from the subject and upstream sections. The convolutional neural network (CNN) was used to capture the spatial characteristic of traffic flow, and the long short-term memory (LSTM) and gated recurrent unit (GRU) neural networks were used to analyze the temporal characteristic of traffic flow. Then, the outputs obtained from these three deep learning methods were integrated to obtain the estimated values of traffic flow parameters. The proposed method was verified using the field data from Hefei city of Anhui province, China and Sacramento of California, United States. The results indicate that the proposed method produces higher accuracy and reliability than existing methods, and reduces the estimation error by 5.72% to 33.29%. The hybrid method can provide high-quality basic data for the intelligent transportation system operation and management.