PMGCN: Progressive Multi-Graph Convolutional Network for Traffic Forecasting

Traffic forecasting has always been an important part of intelligent transportation systems. At present, spatiotemporal graph neural networks are widely used to capture dependencies. However, most use a single predefined matrix or self-generated matrix. It is difficult obtain deeper spatial information by only relying on adjacency In this paper, we present progressive multi-graph convolutional network (PMGCN), which includes attention, convolution, and multi-scale convolution modules. Specifically, new attention that can extract extensive comprehensive dynamic dependence between nodes, in multiple convolutions adopt connections dynamically adjusts each item the Chebyshev polynomial convolutions. addition, time was added from receptive field features. We real datasets predict traffic speed flow, results were compared with variety typical prediction models. PMGCN smallest Mean Absolute Error (MAE), Root Squared (RMSE), Percentage (MAPE) under different horizons (H = 15 min, 30 60 min), shows superiority proposed model.