Setting states - Mcp/0036

[HansOlsson]

The MCP for setting states.
It has been rebased on current master to have a good state for the specification changes.

[HansOlsson]

Should also be ready to enter under evaluation.

[HansOlsson]

I will try to demonstrate at the next meeting

[HansOlsson]

I have had on my agenda to explain this for some time, and finally realized that instead of providing more information, it may be helpful to provide less (i.e., a condensed explanation).

Back-ground:

The basic idea with feed-forward control is that if we have a desired output ($y_{des}$) and the model can be described as a state-less function $y=g(u)$, the solution is $u=g^{-1}(y_{des})$.

Generalization:

If the model has states (and you can observe them) that simplified view changes to $y=g(u, x)$, and the solution is $u=g^{-1}(y_{des}, x)$ (Or more accurately solve $y_{des}=g(u, x)$ for $u$.)

We assume that $\dot{x}=f(x, u)$ - but we don't use that when finding the control signal.

Thus the desired functionality is to be able to evaluate the (inverted) model as a function given the known values of the states, i.e., basically setting the states for any clock tick where we need to invert the model. There is no need to perform any integration of these state, as we will always have new state-values on the next clock tick. That is exactly the functionality in this MCP.

It can be handled with an FMU, but clearly having everything in Modelica seems better.

(Obviously in practice there's also likely feedback control, observers, limiters for the signals etc.)

[HansOlsson]

I believe the links work now. It seems the conference proceedings were moved for some reason.