Crisp: A Spectral-Based Interaction Strategy for Multivariate Time Series Forecasting

Multivariate time series (MTS) forecasting critically depends on modeling inter-variable dependencies, yet existing paradigms face a trade-off: channel-isolation strategies can suffer from information fragmentation in strongly coupled systems, whereas channel-interaction methods often introduce spurious interactions among irrelevant variables. To address this challenge, we propose Coherent Resonance Interaction with Spectral Priors (Crisp). Crisp adopts the principle that effective information exchange should occur only between variables with compatible oscillatory patterns. Concretely, we derive spectral priors in the frequency domain to construct dynamic resonance topologies. With a differentiable, adaptive, and strictly sparse blocking mechanism, Crisp forces attention weights for spectrally inconsistent neighbors to be exactly zero. In addition, we introduce a spectral-gated feature filtering module to refine variable representations using intrinsic spectral characteristics. Extensive experiments demonstrate that Crisp significantly outperforms 20+ baselines. Our code is available at Anonymous GitHub.

Many real-world systems, such as traffic networks, power grids, and weather sensors, record multiple measurements over time. Accurate forecasting requires deciding which measurements should share information. However, sharing too little information can miss useful relationships, while sharing too much can introduce misleading noise. This paper presents Crisp, a forecasting method that allows measurements to interact only when they show compatible repeating patterns. By comparing their underlying rhythms, Crisp keeps helpful relationships and blocks irrelevant ones. It also filters each measurement’s features to reduce noise. Experiments on diverse real-world datasets show that Crisp improves forecasting accuracy and remains robust when data are incomplete.