Neural Network Predictive Controller Design without the plant model but with historical data.

I am trying to model neural network predictive controller using historical data. But with all the examples that I referred, we need plant model to train controller. Is there anyway we can use historical data to model plant?

My many thanks.

Abhishek Guhe

Answer 1

[Arkadiy Turevskiy]

This is applicable to all control design problems in general: if you have plant input-output data and want to create a plant model from them, you can use system identification. MathWorks offers System Identification Toolbox for that. Take a look at this 2 min video to get an idea about how it works.