Predictive control: anticipate rather than correct
Model predictive control (MPC) uses a process model to compute, over a rolling horizon, the set points that optimise a criterion while respecting safety, quality and equipment constraints.
Why go beyond classic regulation
Observed gains are often a few percent of energy and reagents, sometimes more on highly disturbed processes; they must be validated against a baseline.
- Handles several coupled variables at once, where independent PID loops interfere with each other.
- Explicitly includes constraints: temperature, pressure and power limits.
- Anticipates known disturbances: weather, production plan, electricity price.
- Optimises a real cost, such as energy consumed or hourly cost, rather than simple regulation around a set point.
Energy applications
- Utility control: boiler house, chillers, cogeneration, compressed air.
- Managing thermal storage or a battery according to prices.
- Coordinating a heat pump with other heat sources.
- Distillation columns, dryers, furnaces and reactors.
Prerequisites and implementation
MPC needs a quality model, often from a digital twin or identification tests, and reliable measurements. It is installed on top of existing regulation, with a degraded mode that hands control back to base loops on fault. Industrial AI helps update models and forecast loads.
On security, access to the control system must respect IEC 62443 segmentation, and operator validation remains the rule. To start, choose a unit with high energy stakes, measure the baseline for several weeks, then compare by alternating MPC and conventional operation.
Pitfalls to avoid
A half-yearly review of the model and the gains preserves the value of the system.
- A model too simple to represent the dominant non-linearities.
- Badly tuned constraints that make control too cautious.
- No performance tracking after commissioning.
- Lack of operator training, leading them to switch the function off.
Key takeaway: predictive control pays back where processes are coupled, constrained and exposed to changing prices. Begin with one utility or unit, measure rigorously, and expand as operators build confidence in the system.
Going further
- Digital twins for industrial processes
From offline model to connected twin: uses, limits, architecture and success conditions of a process digital twin.
- Industrial AI: use cases and governance
Predictive maintenance, soft sensors, set-point optimisation and operator assistants: deploying AI in plants with a sound method.
- Energy management
ISO 50001, energy audits, sub-metering, PPAs and flexibility: managing industrial energy use sustainably.
- Pinch analysis and heat integration
Pinch method: composite curves, ΔTmin, golden rules and retrofit to reduce a process’s heating and cooling consumption.
- High-temperature industrial heat pumps
COP, temperature lift, refrigerants, waste-heat sources and limits of industrial heat pumps up to 150 °C.
- Industrial thermal storage: sensible, latent, salts
Sensible, latent and thermochemical storage: sizing, losses, industrial uses and coupling with heat pumps and renewables.
- Energy systems optimisation
Pinch analysis, heat recovery, high-temperature heat pumps, thermal storage and predictive control for industrial energy systems.