Error in ode45 (Index exceeds matrix dimensions)

1 次查看（过去 30 天）

显示更早的评论

Arbol 2017-6-1

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/342933-error-in-ode45-index-exceeds-matrix-dimensions

编辑： Walter Roberson 2017-6-5

采纳的回答： Star Strider

在 MATLAB Online 中打开

Can someone please explain what I have to fix for the AF_function_in(t) of my code?

My function is as followed:

    function dx=myeqn(t,x,p)
    global tu;
    function output=AF_function_in(t)
      output=interp1(tu(:,1),tu(:,2),t);
      end
    dx(1,1)= (p(1)/p(2))*(AF_function_in(t)-x(1))- (p(4)/p(2))*(x(1)-x(2));
    dx(2,1)=  (p(4)/p(3)) *(x(1)-x(2));
    end

When I run:

tv=1x225 matrix;
[t,x]=ode45(@(t,x) myeqn (t,x,[0.1 0.2 0.3 0.4],tv,[0 0]));

I get an error of:

Error in myeqn/AF_function_in (line 4)
        output=interp1(tu(:,1),tu(:,2),t);
Error in myeqn (line 6)
dx(1,1)= (p(1)/p(2))*(AF_function_in(t)-x(1))- (p(4)/p(2))*(x(1)-x(2));
Error in @(t,x)myeqn(t,x,[0.1,0.2,0.3,0.4])

采纳的回答

Star Strider 2017-6-1

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/342933-error-in-ode45-index-exceeds-matrix-dimensions#answer_269269

在 MATLAB Online 中打开

First, ‘tu’ has to be an (Nx2) matrix.

Second, do not use global variables! They make your code much more difficult to troubleshoot. Pass ‘p’ and ‘tu’ as extra parameters instead.

Try this:

function dx=myeqn(t,x,p,tu)
output=interp1(tu(:,1),tu(:,2),t);
dx(1,1)= (p(1)/p(2))*(AF_function_in(t)-x(1))- (p(4)/p(2))*(x(1)-x(2));
dx(2,1)=  (p(4)/p(3)) *(x(1)-x(2));
end

and then call it in your ODE solver as:

[t,x] = ode45(@(t,x) myeqn(t,x,p,tu), tspan, x0);

with ‘x0’ being your (2x1) initial conditions vector.

@Star Strider Also, I see that you understand how to use lsqnonlin or lsqcurvefit by following this: https://www.mathworks.com/matlabcentral/answers/254566-how-do-i-fit-coupled-differential-equations-to-experimental-data.

However, I couldn't able to get my code work. Hopefully you can enlighten me. I have tried all possible way. Using this code I made, the function stopped prematurely, and the First-order optimality converge to one value. I use lsqnonlin function to fit my model as followed:

Fp= 0.1; fp=0.1; fis=0.1; PS=0.1;
param0 = [Fp,fp,fis,PS];
fun=@(p) odefnc_1(p,tu(:,1),tu(:,3),tu);
%Then call the lsqnonlin using:
[newparam,resnorm,RESIDUAL,EXITFLAG,OUTPUT,LAMBDA,Jacobian]=...
    lsqnonlin(fun,param0,lb,ub,options);
function diff= odefnc_1(p,texp,x,tu)
F=p(1);
fp=p(2);
fis=p(3);
PS=p(4);
init = [0 0];
[tv,y]=ode45(@myeqn, texp, init);
P=y(:,1);
I=y(:,2);
diff = (I+P)-x;
function dx=myeqn(t,x)
output=interp1(tu(:,1),tu(:,2),t); 
dx(1,1)= (p(1)/p(2))*(output-x(1))- (p(4)/p(2))*(x(1)-x(2));
dx(2,1)=  (p(4)/p(3)) *(x(1)-x(2));
end
end

Star Strider 2017-6-3

在 MATLAB Online 中打开

I just now noticed that you are calling the lsqnonlin function, not lsqcurvefit. The lsqnonlin function is appropriate here. The problem is that I do not see where you are comparing the results of your objective function to your dependent variable. (If you are, and I am not seeing it because I do not know what your dependent variable is, please post that function, and your call to it, specifically.)

If ‘y’ is your dependent variable, I would anticipate that your cost function (that you pass to lsqnonlin to optimise) would be something like this (for lsqnonlin):

fun=@(p) y - odefnc_test(p,tu(:,1),tu(:,3),tu);

Try something like that to see if you get the result you want.

Otherwise, it might be easier to use lsqcurvefit instead of lsqnonlin, since it incorporates the cost function in its code. As I mentioned previously, if your regression hypersurface has many local minima, lsqcurvefit may have a very difficult task in getting the ‘best’ fit global minimum. If that emerges as a problem, I would consider using a genetic algorithm (the ga function, although simple but effective ones are not difficult to write yourself) or the patternsearch function.

Arbol 2017-6-5

Hey @Star Strider, I got it to fit! I just had to pick out a good starting point (param0) for the fit to work! So that is something to note for. Have to pick out a good initial parameter!