Spatiotemporal Forecasting of Traffic Flow using Wavelet-based Temporal Attention.
IEEE Access., 2024
Spatiotemporal traffic forecasting is challenging due to nonlinear dynamics, noise, and long-range dependencies across time and space. This work proposes W-DSTAGNN, a wavelet-based graph neural network that leverages multi-resolution analysis to capture temporal patterns and filter noise effectively. By integrating wavelet decomposition with spatiotemporal attention mechanisms, the model improves representation of complex traffic dynamics.NARFIMA, a hybrid framework that integrates ARFIMA with neural networks while incorporating exogenous variables. The model captures complex dynamics and provides theoretical guarantees, including asymptotic stationarity.
Overview of the W-DSTAGNN Architecture.