Traffic Flow Prediction Algorithm Based on Attention Spatiotemporal Graph Convolution Mechanism

With the continuous advancement of smart city construction, short-term urban traffic flow prediction has become increasingly important. Considering the impact of traffic flow characteristics on traffic flow prediction results, this paper proposes a short-term urban traffic flow prediction algorithm based on an attention spatiotemporal graph convolution mechanism for cities with irregular urban layouts. The algorithm integrates graph convolutional networks and long short-term memory neural networks into an end-to-end network. Firstly, the spatial and temporal characteristics of traffic flow are extracted using graph convolutional networks and long short-term memory neural networks, respectively. Secondly, they are connected with spatial and temporal attention mechanisms to capture the dynamic correlations in the data. Finally, the model is validated using New York City bike data, resulting in reduced values of mean absolute percentage error and root mean square error.