Model-Predictive Control Basics
Model-predictive control repeatedly plans a short future using a model, applies the first step, then re-plans as new data arrives.
Control by looking ahead
Model-predictive control, MPC, is a control strategy that uses a model to look into the near future. At each step it predicts how the system will respond to a range of possible actions over a short horizon, chooses the action sequence that best meets the goals while respecting limits, applies only the first action, and then repeats with fresh measurements. This constant re-planning is what makes it robust.
The optimization at its heart
Each step solves an optimization: minimize a cost that penalizes deviation from targets and excessive actuator effort, subject to constraints that encode physical and safety limits. Because constraints are explicit, MPC naturally keeps the system inside its safe envelope, which conventional controllers do only indirectly.
Why the horizon and re-planning matter
- Looking ahead lets the controller act before a limit is reached, not after
- Applying only the first step and re-planning corrects for model error and disturbances
- Explicit constraints keep every planned action feasible and safe
The cost of MPC
Solving an optimization every control step is demanding, especially with a detailed model. This is the direct reason twins invest in fast surrogates: MPC needs a model it can evaluate many times within one control period. The accuracy of the model sets the quality of the plan, so MPC and the twin are tightly linked.
In a fusion machine
Fusion plants must hold plasma and machine states inside narrow, hard limits while rejecting disturbances, which is exactly what MPC does well. For the Hyperion breeder, MPC can regulate shape and internal profiles against actuator and stability limits; for the burner, it can manage the high-field tandem-mirror configuration. Feeding MPC from the twin's live state estimate closes the loop. See twin-in-the-loop control and coupling the twin to control.