Here we gave a gentle intro to model predictive control (MPC). Some contents comes from Prof. Alberto Bemporad’s slides at IMT Lucca.

Basic Idea

According to Prof. Alberto Bemporad, the core idea of MPC is “Use a dynamical model of the process to predict its future evolution and optimize the control signal.”

Specifically, at time $t$: solve an optimal control problem over a finite future horizon of $N$ steps:

\[\begin{align} \min_u \quad &\sum_{k=0}^{N-1} || y_{t+k}-r(t) ||^2+\rho||u_{t+k}||^2 \\ \text{s.t.} \quad &x_{t+k+1} = f(x_{t+k},u_{t+k}) \\ &y_{t+k} = g(x_{t+k}) \\ &u_{\text{min}} ≤ u_{t+k} ≤ u_{\text{min}} \\ &y_{\text{min}} ≤ y_{t+k} ≤ y_{\text{min}} \\ & x_t = x(t), k =0,...,N − 1 \end{align}\]

After the optimization at each step, only the first optimal move $u^*(t)$ is applied.

At time $t+1$: Get new measurements, repeat the optimization. And so on … A simple illustration of the process can be found in the figure below.

mpc

The advantage of repeated on‐line optimization is FEEDBACK, which is one of the core parts in control.

Constrained Optimal Control

In MPC, we are actually solving a constrained optimization at each step. We consider a linear model: