Fourier Graph Convolution Network for Time Series Prediction

The spatio-temporal pattern recognition of time series data is critical to developing intelligent transportation systems. Traffic flow are that exhibit patterns periodicity and volatility. A novel robust Fourier Graph Convolution Network model proposed learn these effectively. includes a Embedding module stackable Spatial-Temporal ChebyNet layer. development the based on analysis theory can capture features. layer designed traffic flow’s volatility features for improving system’s robustness. represents periodic function with find optimal coefficient frequency parameters. consists Fine-grained Volatility Module Temporal Module. Experiments in terms prediction accuracy using two open datasets show outperforms state-of-the-art methods significantly.