Select, then Balance: Exploring Exogenous Variable Modeling of Spatio-Temporal Forecasting

TL;DR

ExoST presents a backbone-agnostic, two-stage framework for exogenous-variable modeling in spatio-temporal forecasting. The Select stage uses Conditional Embedding and a latent-space gated expert to adaptively extract informative signals from heterogeneous exogenous inputs, addressing inconsistent variable effects. The Balance stage employs a Siamese dual-branch backbone with a context-aware balancer to dynamically fuse past and future exogenous contexts, tackling type imbalance. Extensive experiments across four real-world tasks demonstrate universal improvements, robustness to missing/noisy exogenous data, and favorable efficiency, establishing ExoST as a general, effective approach to integrating exogenous information into ST forecasting.

Abstract

Spatio-temporal (ST) forecasting is critical for dynamic systems, yet existing methods predominantly rely on modeling a limited set of observed target variables. In this paper, we present the first systematic exploration of exogenous variable modeling for ST forecasting, a topic long overlooked in this field. We identify two core challenges in integrating exogenous variables: the inconsistent effects of distinct variables on the target system and the imbalance effects between historical and future data. To address these, we propose ExoST, a simple yet effective exogenous variable modeling general framework highly compatible with existing ST backbones that follows a "select, then balance" paradigm. Specifically, we design a latent space gated expert module to dynamically select and recompose salient signals from fused exogenous information. Furthermore, a siamese dual-branch backbone architecture captures dynamic patterns from the recomposed past and future representations, integrating them via a context-aware weighting mechanism to ensure dynamic balance. Extensive experiments on real-world datasets demonstrate the ExoST's effectiveness, universality, robustness, and efficiency.

Figures (11)

• Figure 1: (a): Comparison between the target-centric view and the proposed exogenous-aware framework for ST forecasting. (b) and (c): A case study of inconsistent variable effects and imbalanced type effects in the AQI-19 dataset.
• Figure 2: Overview of the proposed Exogenous-Aware Spatio-temporal forecasting framework ExoST.
• Figure 3: Visualization of gated expert probabilities.
• Figure 4: Weight distribution across different strategies.
• Figure 5: Left: MRE performance gains of different models on different datasets with and without ExoST. Right: Average relative improvement of different metrics for each model across all datasets (Full results in Appendix \ref{['appendix_universality']} Table \ref{['tab:rq1_3day']}).