i got error in iddata about intersample size..how to fix it?

[Walter Roberson]

It looks to me as if you could get that if you applied diff() to a value of type iddata and the value happened to be empty.

Is it deliberate that you are applying diff to an iddata value? What are the characteristics of that iddata value ?

[KJ DINESHBABU]

i have time deries data of 8 inputs and one output...i create a time series model with this data..when i came to design a non linear MPC..it asking state function to create nlmpc object...Specify Prediction Model for Nonlinear MPC .. the state function is the state derivative function.

so i tried "diff" command to derivate the iddata object (contain 8 inputs and one output)...when i put this iddata to diff command ..i got this error...

what should i give to diff command to complete my nonlinear mpc object..?

why it shows error in "intersample" property?

[Walter Roberson]

When you created the model and processed your data with it, the model should have estimated a number of parameters. Extract those parameters from the model. If you got out one or two row vectors then tf() might be the appropriate call from there, but if you got out a few 2D matrices then ss() might be the constructor.

My suspicion is that you will not need to construct the derivatives yourself: I suspect that the model has the values you need.

However I only did a little work on iddata and that was several years ago, so I might be missing something important.

[KJ DINESHBABU]

thanks sir.

but i have another doubt ...how to design nonlinear MPC controller with neural network timeseries model?

A nonlinear MPC controller does not support:

• Code generation
• Designing controllers using the MPC Designer app.

then how can i design it with historical data using timeseries neural network model?

please explain me how define state function?

[Walter Roberson]

ss() does not care where it gets the numeric matrices. You can create a neural network with a vector of numeric outputs, which you can then extract portions of, and reshape, and pass to ss to create a state space model that you can simulate your model over to measure goodness of. That produces reactions that you can feed in to the neural network.

In theory. There might be easier or better ways.

[KJ DINESHBABU]

for non linear MPC ..it can handle nonlinear model right?

then why need to go for ss()..normal mpc need linearized model..so it convert timeseries neural network into ss()...i tried(normal MPC) it.

nonlinear MPC should accept the neural network directly?