Thank you, please provide link to the estimation and validation data used in the video. Please note the 'linearRegressor' is not defined in Matlab R2018b.. please advise.
Great stuff as usual! Regarding knowledge about the system that can be used, I was wondering whether a saturation (non-linear) function of the output could also be used by these methods, to address in a natural way the limits of the tank (full or empty), for example. It seems that the sigmoid function kind of captures this in a more behaved fashion (continuous function) either way.
great video! and a smooth introduction to NARX models... can you give us some insights about the sufficient richness and persistent excitation of input data to generate a good dataset to train the model?
I have a question. You sair linear models cant handle the offset. However the start point is the initial conditions, and linear models can be initialized at any x0. How is this distintion made in the software?
Thanks for the very planned-out videos you post. I have a miss understanding of a point you rush through it a bit , stability issues of nonlinear systems when predicting output? Why is that occuring? Apart from the external input sequences the system has never seen before. Thanks.
Does anyone have experience with identifing friction in elektro-mechanical systems? I excited my elektro-mechanical system with PRBS as input (current) and recorded the output (velocity). Now i want to identify a model that captures the nonlinearity of the mechanical friction. Any tips?
does anybody else have the following issue: Unrecognized function or variable 'idWaveletNetwork'. ? any ideas what to do about it? i've installed deep learning and system identification toolboxes, but it's not it. would be great if the next video was about nonlinear system identification for MIMO systems
Hi Brian Douglas, your videos helped me a lot. I'm modeling a very slow first order thermal system, but working close to 1200ºC, passing through the specific temperature of a test piece, which I imagine is a disturbance, at least that's what I've seen in the residual graphics. Any advice to achieve a reliable model in this case? Anyway thank you so much for the videos.
Hello Brian, first let me congratulate you for this amazing video, just one question; i am using matlab 2023 and still linearRegressor output the following error: Error using idRegressorSpec/set.Lags The value of the "Lags" argument must be a cell array of unique non-negative index vectors, one vector for each variable.
This is awesome, using intuition from the physical system to improve the model is just amazing
Damn one of the first NAR videos online. Thanks!
he calls himself Brian, we call him GOAT
Finally! Great video!
Thank you, please provide link to the estimation and validation data used in the video. Please note the 'linearRegressor' is not defined in Matlab R2018b.. please advise.
Great stuff as usual! Regarding knowledge about the system that can be used, I was wondering whether a saturation (non-linear) function of the output could also be used by these methods, to address in a natural way the limits of the tank (full or empty), for example. It seems that the sigmoid function kind of captures this in a more behaved fashion (continuous function) either way.
System identification is very close to magic.
great video! and a smooth introduction to NARX models... can you give us some insights about the sufficient richness and persistent excitation of input data to generate a good dataset to train the model?
I have a question. You sair linear models cant handle the offset. However the start point is the initial conditions, and linear models can be initialized at any x0. How is this distintion made in the software?
Can you give a link to this example to play around with it
Thanks for the very planned-out videos you post.
I have a miss understanding of a point you rush through it a bit , stability issues of nonlinear systems when predicting output? Why is that occuring? Apart from the external input sequences the system has never seen before.
Thanks.
Too good, thanks a lot Brian
excellent video . I really appreciate it.
THANK YOUUUUUUUU.
Does anyone have experience with identifing friction in elektro-mechanical systems? I excited my elektro-mechanical system with PRBS as input (current) and recorded the output (velocity). Now i want to identify a model that captures the nonlinearity of the mechanical friction. Any tips?
How can we use Bode Plot to identify our Black Box system?
does anybody else have the following issue: Unrecognized function or variable 'idWaveletNetwork'. ? any ideas what to do about it? i've installed deep learning and system identification toolboxes, but it's not it.
would be great if the next video was about nonlinear system identification for MIMO systems
Hi Brian Douglas, your videos helped me a lot. I'm modeling a very slow first order thermal system, but working close to 1200ºC, passing through the specific temperature of a test piece, which I imagine is a disturbance, at least that's what I've seen in the residual graphics. Any advice to achieve a reliable model in this case? Anyway thank you so much for the videos.
Sir how is SINDy different from it?
but why you're not using sys identification that exists on matlab toolbox and then u just choose the transfert function that gives u the best fit ?
Hello Brian, first let me congratulate you for this amazing video, just one question; i am using matlab 2023 and still linearRegressor output the following error: Error using idRegressorSpec/set.Lags
The value of the "Lags" argument must be a cell array of unique non-negative index vectors, one vector for each variable.
well i figured it out for those who may face the same issue: just name your regressors variables like the nlarx model which are y1 and u1
cool 😁 i thought arx describes the error model H=1/A (y=B/A u + 1/A e)?
thanks)