Abstract:Traffic forecasting is crucial for public safety and resource optimization, yet is very challenging due to the temporal changes and the dynamic spatial correlations. To capture these intricate dependencies, spatio-temporal networks, such as recurrent neural networks with graph convolution networks, are applied. However, traffic forecasting is still a non-trivial task because of three major challenges: 1) Previous spatio-temporal networks are based on end-to-end training and thus fail to handle the distribution shift in the non-stationary traffic time series. 2) Existing methods always utilize the one-hour input to forecast future traffic and the long-term historical trend knowledge is ignored. 3) The efficient and effective algorithm for modeling multi-scale spatial correlations is still lacking in prior networks. Therefore, in this paper, rather than proposing yet another end-to-end model, we provide a novel disentangle-fusion framework STWave + to mitigate the distribution shift issue. The framework first decouples the complex one-hour traffic data into stable trends and fluctuating events, followed by a dual-channel spatio-temporal network to model trends and events, respectively. Moreover, long-term trends are used as a self-supervised signal in STWave + to teach overall temporal information into one-hour trends through a contrastive loss. Finally, reasonable future traffic can be predicted through the adaptive fusion of one-hour trends and events. Additionally, we incorporate a novel query sampling strategy and multi-scale graph wavelet positional encoding into the full graph attention network to efficiently and effectively model dynamic hierarchical spatial correlations. Extensive experiments on four traffic datasets show the superiority of our approach, i.e. , the higher forecasting accuracy with lower computational cost.

Wavelet-Driven Spatiotemporal Predictive Learning: Bridging Frequency and Time Variations

Interpretable and Generalizable Spatiotemporal Predictive Learning with Disentangled Consistency

STONet: A Neural-Operator-Driven Spatio-temporal Network

Temporal Attention Unit: Towards Efficient Spatiotemporal Predictive Learning

Revisiting the Temporal Modeling in Spatio-Temporal Predictive Learning under A Unified View

Predicting High-Dimensional Time Series Data with Spatial, Temporal and Global Information

Wavelet-Temporal Neural Network for Multivariate Time Series Prediction

STWave+: A Multi-Scale Efficient Spectral Graph Attention Network with Long-Term Trends for Disentangled Traffic Flow Forecasting

Spatiotemporal Forecasting of Traffic Flow using Wavelet-based Temporal Attention

SimVP: Towards Simple yet Powerful Spatiotemporal Predictive Learning

Dual-channel Spatio-Temporal Wavelet Transform Graph Neural Network for Traffic Forecasting.

3D Wave Simulation Based on a Deep Learning Model for Spatiotemporal Prediction

Dynamic Spatiotemporal Graph Wavelet Network for Traffic Flow Prediction

Triplet Attention Transformer for Spatiotemporal Predictive Learning

STLight: a Fully Convolutional Approach for Efficient Predictive Learning by Spatio-Temporal joint Processing

Is Single Enough? A Joint Spatiotemporal Feature Learning Framework for Multivariate Time Series Prediction

A Unified Replay-based Continuous Learning Framework for Spatio-Temporal Prediction on Streaming Data

PredRNN: Recurrent Neural Networks for Predictive Learning using Spatiotemporal LSTMs

WEITS: A Wavelet-enhanced residual framework for interpretable time series forecasting

MotionWavelet: Human Motion Prediction via Wavelet Manifold Learning

Forecasting Wavelet Transformed Time Series with Attentive Neural Networks.