Large language model improves scheduling by predicting future jobs
LLM-Based Exogenous World Modeling for Dynamic Scheduling
Artificial IntelligenceComputation and Language
Summary
Scheduling jobs in a factory or workshop is tricky because new tasks keep arriving and may cause delays if not planned carefully. The authors show that a large language model (LLM), a type of AI trained on lots of text, can predict future job arrivals more accurately than traditional methods. They use this prediction to improve job dispatching, deciding which tasks to do next. By updating the predictions when surprises occur and learning from recent errors, their approach reduces delays and lateness in different job scheduling scenarios.
What this means in practice
- •For manufacturing schedulers: Improve job dispatch decisions by forecasting future task arrivals to reduce delays in dynamic factory environments.
- •For logistics planners: Use forecast-updated scheduling to better manage changing delivery jobs and reduce lateness in flexible operations.
Authors
Ni Zhang, Haofeng Yuan, Jianan Zhou, Shiji Song, Zhiguang Cao
Abstract
Dynamic scheduling requires dispatching online arrivals while anticipating future jobs, as poor decisions can compound downstream costs under nonlinear tardiness objectives. We show that a pretrained large language model (LLM) can serve as the exogenous component of a structured world model, forecasting uncertain future job arrivals. Specifically, we exploit an afterstate decomposition, which separates known deterministic shop transitions from uncertain exogenous job arrivals, enabling the LLM to forecast only the latter through a compact belief representation. Building on this, we develop Afterstate World-Model Planning (AWMP), which combines an evolved, belief-conditioned dispatching strategy with short scenario-based rollouts. A surprise-triggered refresh mechanism queries the LLM only when newly observed arrivals deviate sufficiently from previous forecast, and online residual self-calibration further conditions the LLM on its recent forecast errors to handle nonstationary arrivals. Across diverse arrival distributions, AWMP reduces weighted tardiness by 6-21% on dynamic job-shop and 22-46% on flexible job-shop benchmarks, relative to the best of eight classical priority-dispatching rules.