Traffic Flow Prediction in Intelligent Transportation using Spatial-Temporal Graph Convolution Attention Module

The timely and accurate traffic flow prediction is critical for stable and safe Intelligent Transportation System (ITS). Due to the nonlinearity and difficulty of traffic, the existing techniques unable to capture local and global correlations. The Spatial-Temporal Graph Convolution Attention Module (STGCAM) is proposed for capture dynamic correlations among spatial and temporal networks. The PeMS04 and PeMS08 dataset is used which composed 3848 and 1979 detectors. The STGCAM is used to estimate the attention weights for every timestep in sequence of input. The graph convolutional operation collects every node feature with adjacent node when CBAM build adaptive corrections to attain node features. The STGCAM performance is calculated by Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) for PeMS04 and PeMS08 datasets. The STGCAM attains 11.65% and 8.37% of MAPE for PeMS04 and PeMS08 dataset correspondingly when compared to existing mechanisms such as Global Spatio-Temporal Dynamic Capturing Network (GSTDCN), and Dynamic Graph Spatial-Temporal Neural Network (DGSTN).