ConFlux: Multivariate Time Series in Flux, One Unified Forecast in Confluence

Real-world multivariate time series are inherently in flux: different variables evolve asynchronously and interact in complex, time-varying ways, yet accurate forecasting requires these dispersed signals to converge into a single unified prediction. This structural mismatch between dynamic, heterogeneous inputs and a unified forecasting objective poses a fundamental challenge for building general-purpose multivariate forecasting models, especially in zero-shot and large-scale settings. To this end, inspired by the idea that ``\emph{all rivers run into the sea}'', we propose \textbf{ConFlux}, a \emph{general-purpose foundation model for multivariate time-series forecasting} by learning to adaptively integrate cross-channel information under a unified forecasting objective. Specifically, ConFlux first reorders variables to reduce cross-variable entanglement, then aggregates adjacent variables into compact patches that can be processed by a Vision Transformer-style architecture. This design shortens the effective context, reduces attention complexity, and provides a unified token representation for pre-training and downstream tasks. Experiments on 25 public datasets show that ConFlux achieves state-of-the-art performance in zero-shot, fine-tuning, and from-scratch settings, while offering faster inference and lower memory usage.

Predicting the future based on complex, interacting data streams—like weather patterns, traffic networks, or energy grids—is incredibly challenging. These real-world systems are messy and constantly evolving. Traditionally, computer models struggle to process all these interacting variables at once because doing so requires massive amounts of memory and computing power. To solve this, we introduce ConFlux, a new foundational AI model inspired by the concept that "all rivers run into the sea". ConFlux works by first sorting the messy data variables so that highly related ones are grouped closely together. Then, it bundles these related variables into compact "patches," which significantly reduces the amount of data the AI has to process simultaneously. Our research demonstrates that ConFlux not only makes highly accurate predictions across 25 different datasets, but also does so using less memory and with faster processing times than existing models.