AdvancedTopic 69 of 101Signals, Control & Comms
Model predictive control
Optimises predicted future behaviour subject to constraints.
In depth
MPC solves an optimisation each step: predict system response over horizon using a model, minimise cost subject to constraints on inputs and states, apply the first move, repeat. Handles multivariable constrained systems (chemical plants, autonomous driving, building HVAC). Needs computational power and a good model.
Key points
- Receding horizon
- Quadratic programming
- Terminal constraints
ExampleRefinery units use MPC to push operation to constraint limits safely.
Model predictive control optimises future actions subject to constraints, which is valuable in refineries, chemical plants and vehicles.
What is Model predictive control?
Optimises predicted future behaviour subject to constraints.
Why is it important to understand model predictive control?
Model predictive control optimises future actions subject to constraints, which is valuable in refineries, chemical plants and vehicles.
How does MPC work?
At each step it predicts the future with a model, optimises the control moves over a horizon and applies only the first move, then repeats.
Why is MPC popular in industry?
It handles multiple variables and hard constraints directly, which simple PID loops cannot do well.
How is Model predictive control used in practice?
Typical applications include controlling a distillation column with limits on temperature and flow, planning the throttle of an engine to meet emission limits and managing the charge and discharge of a battery under constraints.
What should I learn before Model predictive control?
Study "Lyapunov stability" first (topic 68 of 101 in Signals, Control & Comms). It covers the groundwork this topic builds on.
What are the key points of Model predictive control?
Receding horizon. Quadratic programming. Terminal constraints.
Can you give an example of Model predictive control?
Refinery units use MPC to push operation to constraint limits safely.
What comes after Model predictive control?
The next topic is "Robust control", which builds directly on this one.
How difficult is Model predictive control?
It is classed as Advanced in Signals, Control & Comms (topic 69 of 101). It is an advanced topic. It assumes you are comfortable with the intermediate topics in Signals, Control & Comms.