Function 'fmincon' not supported for code generation

Hello,

im currently trying to implement nonlinear MPC in MATLAB/Simulink using the MATLAB function block.

function u_mpc = MPC(x, u_mpc, u_0, A_ineq, b_ineq, Q_mpc, R_mpc)

costfunc = @(x,u) (x)'*Q_mpc*(x) + (u_mpc)'*R_mpc*(u_mpc);
 
u_mpc = fmincon(costfunc, u_0, A_ineq, b_ineq);

end

The state x is provided by a plant model, the costfunction is a simple quadratic cost with weight matrices, the matrices are taken from the workspace.

Running the Simulink model results in the error: Function 'fmincon' not supported for code generation.

I'm using the 2018b version normally, tried using 2021a but got the same error.

Is there a mistake in the code or is it simply not possible to use fmincon inside a MATLAB function block?

Thanks in advance for any help.

[Walter Roberson]

https://www.mathworks.com/help/optim/ug/fmincon.html#mw_baf6aafe-f2c2-48a7-bd37-0b4aebe61dfd

You need to provide options

[Anuschan Albrecht]

Thanks for your fast answer!

On the help page you linked it says to not load options from a file so I added them inside the MATLAB function block:

function u_mpc = MPC(x, u_mpc, u_0, A_ineq, b_ineq, Q_mpc, R_mpc)
costfunc = @(x,u) (x)'*Q_mpc*(x) + (u_mpc)'*R_mpc*(u_mpc);

options = optimoptions('fmincon','Algorithm','sqp');
u_mpc = fmincon(costfunc, u_0, A_ineq, b_ineq,options);
end

This results in the error: Function 'optimoptions' not supported for code generation.

Answer 1

[Aditya Patil]

fmincon codegen support was added in R2019b. Also, optimOptions is the last argument for fmincon, you should pass other arguments as empty ([]) if you are not using them. For example,

function u_mpc = MPC(x, u_mpc, u_0, A_ineq, b_ineq, Q_mpc, R_mpc)
    costfunc = @(x,u) (x)'*Q_mpc*(x) + (u_mpc)'*R_mpc*(u_mpc);
    options = optimoptions('fmincon','Algorithm','sqp');
    Aeq = [];
    beq = [];
    lb = [];
    ub = [];
    nonlcon = [];
    u_mpc = fmincon(costfunc,u_0,A_ineq,b_ineq,Aeq,beq,lb,ub,nonlcon,options);
end

[Anuschan Albrecht]

Thanks for your answer!

I didn't include them initially because passing empty variables resulted in this error: Inferred size for data 'A_eq' is empty, which is invalid for a block parameter.

Nonetheless, even defining them inside the function as you suggested didn't resolve the original problem of:

Function 'optimoptions' not supported for code generation.

and

Undefined function or variable 'options'. The first assignment to a local variable determines its class.

[Aditya Patil]

Can you share your code?

[Anuschan Albrecht]

function u_mpc = MPC(x, x_ref, u_mpc,u_ref, u_0, A_ineq, b_ineq, Q_mpc, R_mpc)

A_eq = [];
b_eq = [];
lb = [];
ub = [];
nonlcon = [];

costfunc = @(x,u) (x)'*Q_mpc*(x) + (u_mpc)'*R_mpc*(u_mpc);

options = optimoptions('fmincon','Algorithm','sqp');
u_mpc = fmincon(costfunc, u_0, A_ineq, b_ineq, A_eq, b_eq, lb, ub, nonlcon, options);

end

That's the latest iteration of the code inside the MATLAB function block.

[Walter Roberson]

What is the size() of what is being passed in for A_ineq and b_ineq ?

[Anuschan Albrecht]

Input x is a vector of size 10x1.

A_ineq and b_ineq are currently filled with zeros:

A_ineq = zeros(10);

b_ineq = zeros(10,1);

[Walter Roberson]

What are the size() of your various inputs?

I notice that your u_mpc is both an input an and output, and that u_mpc is the result of fmincon() . fmincon() always returns a row vector: is your u_mpc also a row vector ?

Where do you use your input x in your function?

fmincon passes a vector in, but your cost function expects two inputs.

[Aditya Patil]

fmincon codegen support was added in R2019b. The code works fine in R2021a. I recommend updating to latest release.

[Anuschan Albrecht]

Using R2021a the code did indeed work, which is strange because I already tested an earlier iteration on a newer MATLAB version and it still produced errors. It probably had another coding mistake.

Anyways the code is running without errors on R2021a now, thanks for all the help!

[Walter Roberson]

Good point about needing a later release! I normally think about that kind of issue early on, but I missed it this time!