Action conditioned models improve time series planning accuracy
ActTime: Action-Conditioned World Modeling for Time Series
Artificial Intelligence
Summary
Making decisions based on time series data is difficult when future actions can change the outcome. The authors introduce ActTime, a model that uses past data and possible future actions to predict different outcomes. This helps compare potential plans before acting. They show ActTime works better than other methods across many simulated environments like building energy and traffic systems. They also find that being accurate about what actually happened doesn’t always mean good predictions about what would happen under different actions.
What this means in practice
- •For building energy managers: Compare different control actions to optimize energy use based on predicted system responses from historical data.
- •For inventory planners: Evaluate alternative restocking strategies using action-conditioned forecasts to improve supply chain decisions without real-world trials.
Authors
Xinxing Zhou, Xiaojie Yuan, Yanlong Wen, Ming Jin
Abstract
Time-series models increasingly condition on known future inputs. In controlled systems, some of these inputs are actions that can be changed, compared, and optimized before execution. A model must therefore do more than predict the realized future: it must predict how outcomes change under different actions from the same observed history and support reliable comparison of candidate plans. We introduce ActTime, an action-conditioned world model for time series. Over the observed history, ActTime updates variable-specific latent states through an action-conditioned transition and corrects them using observations. It then rolls the same transition forward under candidate actions to generate future trajectories. Training uses only factual trajectories, without paired alternative-action outcomes or planning supervision. We construct a simulator-grounded evaluation suite across building energy, traffic, greenhouse climate, and inventory management, covering factual prediction, intervention-effect prediction, offline decision making, and closed-loop planning. ActTime ranks first in 12 of 16 domain-task settings. Across eight methods, the mean pairwise model-ranking reversal rate is 17.8% between factual prediction and intervention fidelity, and rises to 28.4% between intervention fidelity and decision ranking. These results show that factual accuracy, intervention fidelity, and decision quality are related but not interchangeable.