Channel-wise Retrieval for Multivariate Time Series Forecasting

TL;DR

CRAFT introduces a channel-wise retrieval method for multivariate time series forecasting, improving accuracy by independently retrieving relevant historical segments for each variable using spectral and temporal analysis.

Contribution

It proposes a novel channel-wise retrieval framework with a two-stage pipeline combining time domain pruning and spectral similarity ranking, addressing inter-variable heterogeneity.

Findings

CRAFT outperforms state-of-the-art baselines on seven benchmarks.

It achieves higher forecasting accuracy with efficient inference.

The spectral similarity approach emphasizes dominant periodicities.

Abstract

Multivariate time series forecasting often struggles to capture long-range dependencies due to fixed lookback windows. Retrieval-augmented forecasting addresses this by retrieving historical segments from memory, but existing approaches rely on a channel-agnostic strategy that applies the same references to all variables. This neglects inter-variable heterogeneity, where different channels exhibit distinct periodicities and spectral profiles. We propose CRAFT (Channel-wise retrieval-augmented forecasting), a novel framework that performs retrieval independently for each channel. To ensure efficiency, CRAFT adopts a two-stage pipeline: a sparse relation graph constructed in the time domain prunes irrelevant candidates, and spectral similarity in the frequency domain ranks references, emphasizing dominant periodic components while suppressing noise. Experiments on seven public benchmarks demonstrate that CRAFT outperforms state-of-the-art forecasting baselines, achieving superior accuracy with practical inference efficiency.