Model-based predictive controllers (MPC) with their objective functions and the previousknowledge of the controlled system try to emulate the human thinking in predicting the systembehavior and solving the constrained problems optimally. Model-based Predictive Control isan optimal-control based method for constrained feedback control, where the optimizationproblem is solved at each time step starting from the current state and over a finite horizoncalled the prediction horizon. With respect to all constraints on states, outputs, and inputs onlythe first element of the resulting optimal control sequence is applied to the controlled systemwhile the rest is discarded.
At the following time step, this computation is repeated with a newstate and over a shifted horizon. Therefore, the MPC problem can be addressed as aconstrained optimization problem (with a linear, quadratic, or infinite norm) with recedinghorizon control (RHC) policy. MPC control is considered as the standard and the most popularcontrol approach for the constrained multivariable control problems in process industry andwith some limitations in electric drive applications. Predictive control for power electronics isresearched since the beginning of the 1980’s 62, 63. Its main difference to the well-knownfeedback control schemes is the pre-calculation of the system behavior and consideration ofthis behavior in the control design before a difference between the real value and the referencevalue really occurs. On the other hand, the feedback control only reacts and tries to correct acontrol difference already existing.
Furthermore, in closed-loop feedback control systemsthere must also be a decision, whether the system should be optimized to variations of thereference value or to disturbing influences of the system. Any solution based on linear controltheory or on closed-loop controls can only be a compromise – it never fulfills optimally alldemands at the same time.