【摘要】 针对当前大多数模型对交通流数据空间信息挖掘不充分、无法捕获长序列单元间的信息等问题，提出一种基于时域图卷积神经网络的交通流预测模型。通过阈值权重法重构邻接矩阵，将多层近邻机制嵌入图卷积网络进一步挖掘空间信息；引入时域卷积网络，借助膨胀因果卷积扩大感知野并结合残差网络提取时间信息；运用Dense网络输出结果。利用加州性能评估系统中两个数据集进行评估，其结果表明，该模型性能优于常用的基准模型以及最近提出的多时空图卷积网络模型。更多还原

【Abstract】 Aiming at the problems that most current models do not adequately mine the spatial information of traffic flow data and cannot capture the information between long sequence units, a traffic flow prediction model based on time-domain graph convolution neural network was proposed. The threshold weight method was used to reconstruct the adjacency matrix, and the multi-layer nearest neighbor mechanism was embedded into the graph convolution network to further mine spatial information. The temporal convolution network was introduced to expand the perception field using the expansion causal convolution and the time information was extracted with the residual network. The full connection layer was used to output the results. Using two datasets from the California performance evaluation system, the results show that the proposed model outperforms the commonly used benchmark models and the recently proposed multi-spatiotemporal graph convolutional network model.更多还原