I am currently writing my master thesis and trying to design a controller for my system. However, the system is somewhat unconventional.

It has a large number of inputs and outputs, is static, non-linear and time-invariant. The goal is to control the disturbance. Because it is static, conventional controllers for MIMO systems (NMPC etc.) are only of limited use.

At the moment I am looking for a similar system in another area. Does anyone have an idea in which area such a system exist?

Thanks in advance!

  • $\begingroup$ What do you mean by static system? $\endgroup$ May 25 '20 at 17:19
  • $\begingroup$ @JoshPilipovsky That I assume that the system is so fast that I only have a nonlinear input output mapping. The problems are more the disturbances. I somehow have to control them. $\endgroup$ May 25 '20 at 18:47
  • $\begingroup$ So your I/O is something like $Y(s) = \phi(s)$, where $\phi$ is the nonlinearity right? Are the disturbances delta functions then, since they are so fast? $\endgroup$ May 25 '20 at 20:46
  • $\begingroup$ @JoshPilipovsky Yeah, that is how my function looks like. Actually, it looks like this $F_{x,y}^{(h)} = a^{(h)} e^{- \frac{(x-x_{a})^{2} + (y-y_{a})^{2}} {b^{(h)}} }$ Where $x_{a}$ and $y_{a}$ is my input for one h (I have ~1000 of h). $(x,y)$ is my measurement point (where I have ~1000 as well). $a$ and $b$ are just parameters. I could reduce the problem though to a size of 100x100. One disturbance is rather static. The other is dynamic and I am not even sure how to model it. $\endgroup$ May 26 '20 at 8:41
  • $\begingroup$ Maybe it would be good if you included some extra information about the system. For example, a mathematical model (I suppose you have one) and if you have any info regarding the nonlinearities. $\endgroup$ May 28 '20 at 19:40

Based on your comment, it seems like you are trying to control a dynamical system to track some reference trajectory subject to Gaussian disturbances. Suppose you have some kind of model for the dynamics, then the discrete system can be written as

$$ \begin{align} x_{k+1} &= f_k(x_k) + a_k w_k\\ y_k &= x_k \end{align} $$

where $w_k$ is a Gaussian vector with mean $\mu_k = [x_a^{(k)}, y_a^{(k)}]^\intercal$ and covariance $\Sigma_{x_k} = \frac{b^{(k)}}{2} I_2$. This is then a stochastic Control problem and there are various methods to create a stabilizing controller just check out the literature.

  • $\begingroup$ Ok, thanks. I will have a look. Though, I have one disturbances which is not stochastic and very strong. It causes that the outputs of some variables h fade partly or completely. Is stochastic control still a solution then? $\endgroup$ May 27 '20 at 9:42

I think there are many systems similar to your system.

A very common way to write a system with a nonlinearty is by writing it as Luré Type System, where the nonlinearity is in the feedback loop $$ \begin{align} \dot{x}(t) &= Ax(t) +Bw(t)\\ z(t) &= Cx + Dw(t)\\ w(t) &= \Delta(t,z) \end{align}$$ where $\Delta$ is the nonlinearty.

Depending on the type of nonlinearty, you can apply all kinds of controller synthesis such as $H_\infty$, $H_2$ and QP.

For example, you can create a sector condition for the nonlinearty and use the Circle Criterion to check if the system is absolutely stable. A common nonlinearty to use in this framework is a saturation.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.