Phys-JEPA: Physics-Informed Latent World Models for Multivariate Time-Series Forecasting

2026-06-14

AI Summary

This work addresses the challenge of ensuring physical consistency at the latent state level in multivariate time series forecasting, a limitation in existing methods that often leads to uninterpretable predictions. The authors propose Phys-JEPA, a novel architecture that uniquely shifts physical constraints from the output space into the latent state space, explicitly embedding known physical laws into both the latent representations and their dynamical evolution. Unmodeled dynamics are captured via residual modules, enabling the model to retain fidelity to physical principles while remaining data-driven. By integrating joint-embedding predictive architecture (JEPA) with physics-guided representation learning, Phys-JEPA achieves interpretable modeling that respects underlying physical structure. The method attains state-of-the-art performance across multiple benchmarks, including Jena Climate, Traffic, and Electricity datasets—evidenced by a temperature prediction MSE of 0.01831 at horizon H=24 on Jena—demonstrating its effectiveness and generalizability.

Abstract

Multivariate forecasting in physical systems requires models that predict coupled temporal variables while preserving meaningful state evolution. Deep forecasters can fit temporal correlations, and physics-informed models can regularize predictions with scientific constraints, but these directions are often connected only at the decoded-output level. As a result, the hidden predictive state that generates future trajectories may remain statistically useful but physically unstructured. We introduce Phys-JEPA, a physics-informed joint-embedding predictive architecture for multivariate time-series forecasting. Phys-JEPA learns a latent world model in which predictive states are decomposed into physical and residual components, and physical consistency is imposed directly on latent states and latent transitions rather than only on decoded forecasts. This formulation uses known physical variables to organize the representation space while retaining residual capacity for unresolved dynamics. On Jena Climate 2009--2016, Phys-JEPA reduces aggregate MSE from 0.12482 to 0.12273 and temperature MSE from 0.01892 to 0.01831 at H=24. On Traffic, full Phys-JEPA improves aggregate MSE over the supervised baseline across all tested horizons, reducing H=192 MSE from 0.800784 to 0.773873. On Electricity, the best variant depends on horizon: static latent consistency is strongest at H=24 and H=48, while full Phys-JEPA gives the best aggregate and target-variable MSE at H=192. These initial results suggest that moving physics-informed learning from output space to latent predictive state space is a promising direction for interpretable temporal world models.

Problem

• multivariate time-series forecasting
• physics-informed modeling
• latent world models
• physical consistency
• predictive state

Innovation

• physics-informed learning
• latent world model
• joint-embedding predictive architecture
• multivariate time-series forecasting
• physical consistency