How can I use fmincon to set up a nonlinear model predictive control for MIMO plants.

I want to minimise the same cost function defined in the MPC toolbox. I want to use a nonlinear black box model which is a comercial simulator. I have an interface between Matlab and the simulator in order to read the outputs and to apply the control signals.

My confussion is about how to define several inputs and output constraints in the fmincon function handle.

best regards

Answer 1

[HIMOUR]

Hello, In the link below some complete examples about how to use "fmincon" in predictive control, you'll find a matlab function "nmpc" that uses fmincon, and some of it's examples. The problem for me is how to transmit reference signal

http://numerik.mathematik.uni-bayreuth.de/~lgruene/nmpc-book/matlab_nmpc.html

examples of nmpc with fmincon

[Jhonatan Camacho]

Hi HIMOUR,

Thanks for your answer. I think the NMPC routine proposed by lars grune is a good start point for use the fmincon.

I have the following appreciations:

1.) The authors divide the code in three sections: update states, solve optimal control and apply control. This scheme is easy to adapt for black box models because the use of user defined functions.

2.) The objective function is based on minimize tracking error for system states. I think, it is possible to adapt the runningcosts user function in order to transmit the reference signal. However, it would be grate if is possible to minimize cost similar to MPC toolbox: http://www.mathworks.com/help/mpc/ug/qp-matrices.html

3) I implement a code using the NMPC routine as template. I approximate the delta u using the diff command. I use a simple sumation over the prediction horizon of the error signal and input cost.

4.) The results for me were very bad. I think it is because the fmincon function use backward/forward difference scheme to estimate the derivatives.

Thanks in advance for your answer. If you find other example using fmincon for MPC it would be interesting.