基于时空卷积神经网络的数据缺失交通流预测

Traffic Flow Prediction with Missing Data Based on Spatial-Temporal Convolutional Neural Networks

针对数据连续缺失情况下交通流预测精度下降甚至失效的问题,提出了一种时空卷积神经网络预测模型,根据横向分布的时间相关性和纵向分布的空间相关性,构建路网交通数据时空矩阵,引入掩码矩阵标记数据的缺失状况,利用卷积操作提取路网中各检测器之间隐含的非线性关联,建立当前时刻与未来交通状态的映射关系,实现数据缺失情况下的交通流预测。使用公开数据集,在3个时间尺度上的验证结果表明,所提出的模型在平均误差和预测精度两个方面均优于长短期记忆网络、门控循环单元、扩散卷积神经网络和图马尔可夫网络模型,在交通数据随机缺失和连续缺失两种情况下,均表现出了良好的稳定性和健壮性。

In order to improve the accuracy of traffic flow prediction under the condition of continuous missing data,a spatial-temporal convolutional neural network prediction algorithm is proposed.The spatial-temporal matrix of road network traffic data is established by considering the spatial adjacent relationship of horizontal distribution and the time dependence of vertical distribution.A mask matrix is introduced to represent the missing situation of data.The convolution operation is used to mine the implicit nonlinear correlation between the upstream and downstream detectors,establish the mapping relationship between the current and future traffic states,and realize the traffic flow prediction under the condition of continuous data absence.The experimental results show that the accuracy of traffic flow prediction is better than those of LSTM,GRU and GMN models in the absence of continuous traffic data.The model shows good stability and robustness,and improves the accuracy of traffic flow prediction in the absence of data.