How can I apply bayesopt to my Simulink model?

Dear all,

I'm doing a thesis about semi-active suspension, systems that can only change the viscous damping coefficient of the shock absorber. The case that I'm studying is with magneterheological damper, which is a damper filled with magneterheological fluid. This kind of fluid can change viscosity changing the magnetic field.

I did in Simulink the quarter car model, now my purpose is to find the best value of my control variable "u_MR" (constant value), that maximize the passenger comfort. The comfort is evaluated with the index of performance "J", calculated as the RMS of the (vertical) acceleration, which is the final output of my model.

Thus, the problem is to find the optimal value of uMR that minimize the index "J" using the Bayesian Optimization and in Matlab the function bayesopt().

I'm finding problems writing the objective function, input of the bayesopt function...

The other input of bayesopt() is the variable:

uMR_opt = optimizableVariable('uMR',[0,f_max]);

I have the acceleration, which is a vector, result of the simulation that I can extract in Matlab and than I can calculate the index "J" (which is a single value, e.g. J = 1.6 m/s^2), but than I don't know how to write the objective function, related to this.

out = sim("quarter_car_model.slx");
%extracting the acceleration from simulink
output_xs2punti = getElement(out.yout, 'xs_2punti');
xs_2punti = output_xs2punti.Values.Data(:);
    
J_sim = rms(xs_2punti); %index of performance

Any suggestion on how to do the objective function?

Thanks to all

Answer 1

[Alan Weiss]

As the documentation states, the objective function accepts a 1-by-D table of values, where D is the number of variables. So you need to write your objective function like this:

function J_sim = fun(x)
xval = x.uMR; % x is a table, xval is a scalar represending the uMR value
% Call your simulation with xval as the uMR value. Then
output_xs2punti = getElement(out.yout, 'xs_2punti');
xs_2punti = output_xs2punti.Values.Data(:);
    
J_sim = rms(xs_2punti); %index of performance
end

Alan Weiss

MATLAB mathematical toolbox documentation

[Elia Bontempelli]

Thank you very much for your quick response.

I can't understand what is the input argument of "fun", what represent "x"?

uMR_opt = optimizableVariable('uMR',[0,f_max]);
results = bayesopt(fun(),uMR_opt,'AcquisitionFunctionName','expected-improvement');

When I call this function from my script what do I have to pass to "fun"?

Thank you

[Alan Weiss]

As the documentation states, fun should be a function handle to your objective function. If you write your objective function the way I just showed, your bayesopt call would be

results = bayesopt(@fun,uMR_opt,'AcquisitionFunctionName','expected-improvement');

If you need to pass extra parameters or data to your objective function, such as a workspace variable or structure params, then your function would be

function J_sim = fun(x,params)
% code goes here
end

and you call it like this:

results = bayesopt(@(x)fun(x,params),uMR_opt,'AcquisitionFunctionName','expected-improvement');

See Passing Extra Parameters.

Alan Weiss

MATLAB mathematical toolbox documentation

[Elia Bontempelli]

It works perfect, thank you very much!!