基于注意力机制的CNN-BiGRU的短时交通流预测

Short-Term Traffic Flow Prediction of CNN-BiGRU Based on the Attention Mechanism

摘要: 短时交通流预测是实现道路交通管理和指导的基础，对道路通行能力、交通安全等具有重要的实际意义。提出一种基于注意力机制的卷积神经网络和双向门控循环单元神经网络相结合的深度学习交通流预测模型。采用卷积神经网络（CNN）提取交通流的空间特征。利用双向门控循环单元（BiGRU）从前向和后向传播中都获取信息，充分提取交通流的时间相关特征。引入注意力机制，将交通流最重要的特征得到了最大限度的保留，以此来提升CNN和BiGRU网络的特征提取效果，达到提升模型的预测精度的目的。在真实数据集上的对比实验和消融实验结果都表明：基于注意力机制的CNN-BIGRU的短时交通流预测模型能够有效捕获交通流的动态时空特征，具有良好的预测性能。

Abstract: Short-term traffic flow prediction is the basis of road traffic management and guidance, and has important practical significance for road capacity and traffic safety, and so on. A short-term traffic flow prediction based on CNN-BiGRU-Attention (CNN and BiGRU network based on attention mechanism) model is proposed. The Convolutional Neural Networks (CNN) have been employed to extract the spatial features of the traffic flow. The Bidirectional Gated Recurrent Unit (BiGRU) is exploited to obtain information from both forward and backward propagation, and then fully capture tem-poral dependencies of the traffic flow. The attention mechanism is employed to maximize the retention of the most important features of the traffic flow, so as to improve the feature extraction effect of CNN and BiGRU networks, and achieve the purpose of improving the prediction accuracy of the model. The results of comparison experiments and ablation experiments on the real data set show that the short-term traffic flow prediction model of CNN-BIGRU based on the attention mechanism can effectively capture the dynamic spatio-temporal characteristics of traffic flow, and has good prediction performance.