System Identification of MIMO
Show older comments
Hi guys I'm working on a project of quadrotor control.


I've done the first principles modelling using SimMechanics, and now I would like to compare the plant model with a model derived from input/output data from the real life.
My question is about how to use System identification when you have more than one input and output, does I need to input random values at all inputs at the same time? Ore try to create models for each output?
Thanks in advance!
Answers (2)
Sebastian Castro
on 28 May 2014
Edited: Sebastian Castro
on 1 May 2019
1 vote
If you already have a model in SimMechanics, however, and you just want to match parameters in that model to your data, I'd recommend using the parameter estimation capabilities in Simulink Design Optimization (if this product is available to you). Here is an example of parameter estimation for an inverted pendulum model in SimMechanics .
3 Comments
Daniel Abara
on 19 Aug 2021
Edited: Daniel Abara
on 19 Aug 2021
@Sebastian Castro thank you for your answer. However, OP asked if you need to give inputs to the system all at once, or if you need to do it separately (channel by channel).
I have a similar problem. I have designed a controller to stabilize a quadrotor. Now I want to identify the entire system as a whole, that is from xd,yd,zd,roll_d,pitch_d,yaw_d, to the measured x,y,z,roll,pitch,yaw. So I can use the model as an agent in a co-operative controller design. So similar to OPs question, do I obtain the data, channel by channel, or do I give setpoints such as PRBS (for example), to all the inputs at the same time, while measuring the outputs.
Thanks for the link you provided above. However, it does not answer that question but it very well explains how to manipulate the data and perform MIMO identification, thank you.
Chandan Sinha
on 24 Jan 2023
@Daniel Abara I am in the same boat as you. I am doing system identification of a hexacopter platform. Do let me know how you ended up doing it for your drone. Should we give channel by channel excitation (one at a time) and obtain a transfer function (or related state space/nonlinear) model for each path from input to output?
Branislav
on 15 May 2023
Still no answer to this question? Such an important topic, however basic answers are missing.
Osama Zekry
on 11 Aug 2023
0 votes
Hi Leonardo,
Did you find a solution for this problem? Please, provide me with any advice. Your response is much appreciated.
Categories
Find more on Time and Frequency Domain Analysis in Help Center and File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!