Adaptive method improves forecasting for varied spatial temporal patterns
AdaST: Adaptive Coupling for Spatial-Temporal Forecasting
Artificial Intelligence
Summary
Forecasting things that change over space and time, like traffic or weather, is tricky because sometimes the data depends more on time, sometimes more on location, or both. The authors found that current methods struggle because they assume these dependencies are always strong and balanced. They created AdaST, a system that breaks down the data into parts based on how time and space relate, then combines these adaptively to make better predictions. Tests show AdaST works better than previous approaches by adjusting to the actual patterns in data.
What this means in practice
- •For traffic operation centers: Improve traffic flow predictions by adapting models to how traffic patterns change over time and locations.
- •For energy grid operators: Enhance short-term energy demand forecasting by dynamically adjusting how spatial and temporal data influence predictions.
Authors
Zhenyu Lei, Chenghao Liu, Yushun Dong, Qi R. Wang, Jundong Li
Abstract
Spatial-temporal (ST) forecasting underpins many real-world systems such as traffic, climate, and energy networks. While existing methods implicitly assume strong spatiotemporal coupling, we observe that real-world ST data exhibits distinct coupling regimes, ranging from temporal-dominated and spatial-dominated to strongly coupled patterns. This mismatch causes current models to suffer from spurious dependencies and degraded performance when one correlation dominates. To overcome this limitation, we aim to dynamically modulate spatial and temporal modeling based on the data's inherent coupling structure. However, three key challenges exist: unknown coupling structure, heterogeneous coupling dynamics, and suboptimal spatial modeling. We propose AdaST, an adaptive ST forecasting framework that tackles these challenges through a decompose-recompose paradigm. AdaST factorizes inputs into components capturing different coupling patterns using heterogeneity-aware experts. Each component is processed by role-aligned modules, and a correlation-informed adaptive recomposer integrates them for final prediction. Extensive experiments confirm that AdaST significantly outperforms state-of-the-art baselines, validating the necessity of an adaptive approach.