Welcome to ControlSystemIdentification.jl Discussions!

[baggepinnen]

Feel free to ask usage questions or questions about system identification in general.

[DionysisPap]

Hi, i wanted to ask what i am doing wrong when i am trying to add this specific package to VS Code.
I have the last version of Julia v1.7.3 and i have the extension to VS Code.
But when i go to add the package i have an error that say "9 dependencies errored". And the errors are about:

DSP
MonteCarloMeasurements
DiffEqBase
LowLevelParticleFilters
DiffEqCallbacks
OrdinaryDiffEq
DelayDiffEq
ControlSystems
ControlSystemIdentification

I dont know if i have to add the specific package with some alternative way. Thanks

[DionysisPap]

I solve it by adding the version 2.3.0, that was successfully compliled and then updating to the newer version 2.4.0. I can't figure out what was the problem exactly.

[baggepinnen]

Hello, I'm happy you solved the problem!

[m-bossart]

Hello,

I'm looking to perform System ID on data which comes from multiple independent experiments of the same system. Is there a way to deal with this type of data and fit a model to all of the data. Something similar to this functionality in MATLAB:
https://ch.mathworks.com/help/ident/ug/dealing-with-multi-experiment-data-and-merging-models.html
Thanks!

[m-bossart]

• My plan is to formulate the problem such that the initial conditions are always 0.
• The system has one input and one output. The input and output are both complex numbers.

[m-bossart]

Thank you for the example, that is very helpful.

[m-bossart]

My understanding is that arx is a reasonable starting approach for my system (although maybe not the best)

[baggepinnen]

I should warn you that this package has not been tested with complex data, and there are likely places in the code where the adjoint operator a' is used instead of transpose in a way that would be incorrect for complex data. I looked through the arx-related code and didn't see any obvious problems, but be sure to validate your results thoroughly :)

Since it's a SISO system, arx should be a reasonable starting point. arx is known to favor a high-frequency fit, but this can be somewhat mitigated by using total least-squares as estimator instead of standard least squares. See the following resources for more info on this

• https://nbviewer.org/github/JuliaControl/ControlExamples.jl/blob/master/identification_arx.ipynb
• https://baggepinnen.github.io/ControlSystemIdentification.jl/dev/tf/#Usage-example:
• https://nbviewer.org/github/JuliaControl/ControlExamples.jl/blob/master/identification_robust.ipynb

[baggepinnen]

I made a new release where the support for multiple datasets is now built into arx and era