Model Predictive Control

Model Predictive Control (MPC) is the receding-horizon optimisation framework in which, at every control instant, the operator solves a finite-horizon optimisation problem using the current system state and forecasts of disturbances (load, weather, prices), applies only the first control action, advances one step, then repeats. The model is typically a MILP, QP or NLP encoding the system dynamics and constraints; the horizon ranges from minutes to days depending on the application. MPC is the standard tool for microgrid operation (Parisio, Rikos and Glielmo 2014; Olivares et al. 2014), building HVAC control, chemical-process control, supply chain and autonomous vehicles. Variants: deterministic MPC (single forecast), stochastic MPC (sample-average over scenarios), robust MPC (worst-case under uncertainty set), economic MPC (objective is operating cost rather than tracking error), distributed MPC (multiple subsystems coordinate via ADMM or dual decomposition). Strength: handles constraints and uncertainty in closed loop; weakness: per-instant computation time must fit within the control period.