Rethinking Urban Mobility Prediction: A Super-Multivariate Time Series Forecasting Approach
Published:
2023.12
Our contributions to the long-term urban mobility prediction challenge using SUMformer are as follows:
- We present a novel super-multivariate perspective on grid-based urban mobility data. Through this approach, we are able to utilize general multivariate time series forecasting models to achieve long-term urban mobility predictions.
- We present the SUMformer: a Transformer model designed to leverage temporal, frequency, and cross-variable correlations for urban mobility forecasting. Notably, it stands out as one of the few Transformer models that explicitly taps into and harnesses cross-variable correlations across every channel and grid for urban mobility prediction.
- Experiments demonstrate that SUMformer surpasses state-of-the-art methods across five real-world datasets. We emphasize the significance of the super-multivariate perspective, explicit cross-variable correlation modeling, and frequency information for achieving optimal performance.