World modeling improves time series forecasting with multimodal data
WorldTS: World Modeling for Multimodal Covariate-aware Time Series Forecasting
Machine Learning
Summary
Forecasting future values in time series usually means predicting future numbers based only on past numbers. But these past observations only show part of what’s going on behind the scenes, and outside factors can also influence the future. The authors propose WorldTS, a method that first learns hidden states influenced by various types of external data and then uses those states to predict future observations. This approach helps make better forecasts by understanding both the hidden dynamics and external influences. They tested it on many real datasets to demonstrate its effectiveness.
What this means in practice
- •For data science teams: Improve forecasting models by integrating multiple external data types to better capture underlying system dynamics in complex time series.
- •For supply chain planners: Enhance demand predictions by including various external signals such as weather and economic indicators directly into the forecasting process.
Authors
Yuhan Zhu, Xiangfei Qiu, Hanyin Cheng, Wangmeng Shen, Chenjuan Guo, Bin Yang, Jilin Hu, Christian S. Jensen
Abstract
Time series forecasting is typically framed as learning a direct mapping from historical to future observations in the observation space. However, sequences of observations generally provide only a partial view of the dynamics of the underlying system, with future observations being shaped by latent dynamics. Recent latent-space forecasting methods thus achieve improved performance by predicting future observations from latent-space representations of historical observations rather than directly forecasting future observations in the observation space. Next, while future observations are also shaped by external factors, how to incorporate external, often multimodal, information into forecasting, so that it can shape latent-state formation and evolution directly, remains underexplored. We propose WorldTS, a world-modeling based forecasting framework that integrates multimodal covariates directly into the forecasting to further improve forecasting performance. Specifically, WorldTS employs a two-stage training strategy. First, it learns forecasting-relevant latent state dynamics conditioned on multimodal covariates, yielding encoded future states. Next, the learned state dynamics are frozen, and an observation decoder is trained to map the predicted future states back to future observations. Extensive experiments on 21 real-world datasets offer insight into WorldTS and its effectiveness.