How can I autotune some parameters based on a Simulink model?

Hello,

Target: Tune a PI controller (Kp and Ki parameters) based on a desired behaviour (y_desired). Problem: The tuned parameters differ too much from the desired ones. The number of iterations is way lower than the one 'MaxIter'.

Working files: 1. simulation_model.mdl 2. autotune.m

The simulation model has a "To File" block in order to compare with the autotune file. This "To File" block saves data as "simulation_datafile.mat" and the used variable name is "simulation_data".

File: autotune.m function [Kp,Ki] = autotune

% Desired plant. PI tuning is done to achieve this behavoiur. 
ghi = 1/sqrt(2);
wn = 4/1;
G_ideal = tf(wn^2,[1 2*ghi*wn wn^2]);
[y_desired,t_desired]=step(G_ideal,0:0.1:15);

% Configure lsqcurvefit
i=0;                            % Iteration counter
pi0 = [100 100];                % Set initial values for the PI controller
options = optimset('Display','off','MaxIter',1e3); %lsqcurvefit
lower_boundary = [0 0];         % PI values have to be positive values
upper_boundary = [1 1].*1e10;
pi = lsqcurvefit(@model, pi0,t_desired, y_desired, lower_boundary ,upper_boundary , options);
Kp = pi(1); 
Ki = pi(2);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    function F = model(pi, t_desired) % Why does t_desired need to be here?
      Kp = pi(1);
      Ki = pi(2);
      i=i+1;

        sim('simulation_model','SrcWorkspace','Current'); % Run model
        load simulation_datafile.mat;
        t_sim =  simulation_data(1,:);
        y_sim =  simulation_data(2,:);
        % Plot result - Show evolution
        figure (1);
        clf;                      
        hold on;
        grid on; 
        title (['iter: ', num2str(i), '   Kp: ', num2str(Kp), '   Ki: ', num2str(Ki)]);
        plot(t_sim,y_sim,'-b',t_desired,y_desired,'-r')
        F =  y_desired- y_sim';           % Target function
      end
  end

Thanks in advance.

Jon Kepa

Answer 1

[Ryan G]

Simulink already has a PID Autotuning capability. Since it looks like you are tuning to a step this should be easy.

You can use simulink design optimization to input the signal and have it tune the gains to match.