基于多层次注意力机制的交通节点分类

不同交通节点对交通需求和对交通状况的贡献不均衡,在进行道路修建、交通设施设置的时候需要考虑交通节点的交通需求,因此在城市中对交通节点进行分类,有助于交通规划的实施。本文构建基于多层注意力机制的深度网络学习模型,对交通图的节点进行嵌入分类。该模型由2层神经网络组成,第1层通过多层注意力计算节点的低维嵌入,第2层将第1层的输出节点嵌入通过聚合操作和softmax激活输出节点的标签概率,采用交叉熵损失函数训练模型参数。最后将该模型应用于某城市交通流量图,证明了其对于节点分类的有效性。

The Classification of Traffic Node Embedding Based on Multi-layer Attention Mechanism

The contribution of different traffic nodes to the traffic demand and traffic condition is not balanced.It is necessary to consider the traffic demand of traffic nodes in the process of road construction and traffic facilities setting.Therefore,classifying traffic nodes in cities is helpful to the implementation of traffic planning.A deep network learning model based on multi-layer attention mechanism is constructed to embed and classify the nodes of traffic map.The model is composed of two layers of neural networks.The first layer embeds the low dimensional attention computing nodes,and the second layer embeds the output nodes of the first layer into the label probability of the output nodes activated by aggregation operation and softmax.The cross entropy loss function is used to train the model parameters.Finally,the model is applied to a city traffic flow chart to prove its effectiveness for node classification.