Research on Urban Traffic Flow Prediction Model Utilizing Heterogeneous Data Fusion

Given the diverse nature of data tangentially associated with traffic flow, shallow fusion in traffic flow prediction proves challenging to enhance predictive accuracy and may diminish the performance of the prediction network. To tackle the aforementioned issues, this study presents a model for predicting traffic flow founded on heterogeneous data fusion. For the primary traffic-related data, a blend of graph convolutional neural network (GCN) and gated recurrent unit (GRU) is employed. Here, GCN predominantly captures the spatial characteristics of the nodes, while GRU focuses on capturing temporal dependencies. Utilizing the network embedding model for indirectly related traffic data involves spatial and temporal feature embedding. This approach enables the network to handle a broader spectrum of urban traffic data, thereby enhancing the accuracy of the deep learning model for the prediction of traffic flow. Experiments conducted on real-world datasets validate the efficacy of embedding fusion of external traffic factors in comparison to traditional traffic prediction methods in the prediction task.