基于GCN-BiLSTM的短时交通流预测模型

Short-term Traffic Flow Prediction Model Based on GCN-BiLSTM

文中提出图卷积网络(GCN)与双向长短时记忆神经网络(BiLSTM)组合短时交通流预测模型.利用图卷积网络提取路网拓扑结构解决拓扑关系问题,提取路网间的空间特征,利用双向长短时记忆神经网络用于学习交通数据的动态变化以获取时间相关性,融合GCN-BiLSTM模型同时考虑路网时空关系实现交通流预测.结果表明:文中提出的方法能更好地适应在不同交通流特性条件下的交通流,工作日和周末的预测偏差相较于经典算法降低12.24%和13.20%.

A short-term traffic flow forecasting model combining graph convolution network(GCN)and bidirectional long-term memory neural network(BiLSTM)was proposed.The topological structure of road network was extracted by graph convolution network to solve the topological relationship problem and extract the spatial characteristics between road networks.The bidirectional long-term and short-term memory neural network was used to learn the dynamic changes of traffic data to obtain the time correlation,and the GCN-BiLSTM model was fused to realize the traffic flow prediction considering the time-space relationship of the road network.The results show that the method proposed in this paper can better adapt to the traffic flow under different traffic flow characteristics,and the prediction deviation of working days and weekends is reduced by 12.24%and 13.20%compared with the classical algorithm.