Autonomous predictive-adaptive simulation for operations support

This paper describes a simulation system that monitors operations on a production floor, periodically creating a model of those operations, and running a simulation that predicts the next several shifts' worth of events, providing operators with new predictive analysis capabilities. As a set of procedures is carried out in the model and the real world, deviations are introduced by the variations between the expected activities and the actual occurrences. More deviations arise from explicit adaptations undertaken by operations staff in response to already-observed anomalies. With each cycle, those deviations are integrated into the model, heuristics are applied to estimate the likely future course of events, and after a simulation run, a new set of predictions is generated from that model, and it compares the new predictions with the last run's predictions. Differences between the pre-loop prediction and the post-loop prediction serve to indicate whether the situation is improving or degrading.