A Traffic Data Imputation Method Based on Spatial-temporal Synchronous Graph Recovery Neural Network

Due to comprehensive considerations regarding demand and cost control in traffic management, the performance indicators for checkpoint sensors can differ depending on the section of road. Consequently, some traffic data collected at checkpoints may be incomplete, which complicates data analysis. To address this issue, we propose using a neural network based on embedding synchronised spatio-temporal map data. Firstly, we designed a trajectory vectorisation algorithm using word embedding techniques, modelling the road network with vehicle trajectories to capture spatial correlations among road checkpoints. Secondly, we constructed a spatio-temporal synchronisation graph recovery neural network (STSGRN) that uses graph convolutional networks (GCNs) and gated recurrent units (GRUs) to account for the spatio-temporal characteristics of traffic data and fill in missing data across multiple dimensions. Finally, we developed an attention mechanism module for the spatio-temporal graph to extract dynamic dependencies at the spatio-temporal level. This architecture enhances the flexibility of the STSGRN in processing complex data with missing values. Experimental results demonstrate that the model effectively identifies spatial and temporal correlations in traffic data, enabling the accurate imputation of missing values. Compared to existing methods, it demonstrates significant improvements and superior generalisation performance.