optimization error ,when i used fmincon for three variables,getting same intial guess value in result

Question

RAHUL KUMAR 2019-8-19

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/476634-optimization-error-when-i-used-fmincon-for-three-variables-getting-same-intial-guess-value-in-resul

评论： RAHUL KUMAR 2019-8-19

clear all
close all
xo=[0.0011,0.0049,0.36];    % assumptions
%nvars=5;
A=[];
b=[]; 
Aeq=[];
beq=[];
lb=[0.001,0.002,0.01];   %lower bound
ub=[0.01,0.08,0.5];    %upper bound
nonlincon = @(X)constraint_new(X);    % calling constraint function
Fitnessfun=@(X)weight_testvariable(X);  %calling objective function
options = optimoptions(@fmincon,'Display','iter-detailed','Algorithm','sqp','MaxIterations',1500)
[X,fval]=fmincon(Fitnessfun,xo,[],[],[],[],lb,ub,nonlincon,options)

baically i got same value of my intial guess value,not all only 1st and 3rd variables.i know these value depend on constraint and objective function

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Walter Roberson 2019-8-19

Can you post the code for constraint_new and weight_testvariable ?

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Walter Roberson 2019-8-19

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/476634-optimization-error-when-i-used-fmincon-for-three-variables-getting-same-intial-guess-value-in-resul#answer_388122

在 MATLAB Online 中打开

function [fval]=weight_testvariable(X)
[...]
X=[0.001;0.006;0.4];

You are ignoring the input and always calculating the same thing.

If you were to not ignore the inputs, then:

thick_i=0.0014-X(1);    
thick_o=0.0075-X(2);

Those would become 0 when X(1) become 0.0014 and 0.0075 . Your fval calculation multiplies the first term by thick_i and the second term by thick_o so the two terms can both become exactly 0 . Is it possible to do better? Perhaps, since weight_testvariable(ub) comes out negative.

So then let us try

X = sym('x', [1 3]);
[c,ceq] = constraint_new(X)
 
c =
 
[ x1 - 7/5000, x2 - 3/400, 27/5000 - x2]
ceq =
         -3.24490449304822

Woah! Why did ceq, the equality constraint, come out as a value that is independent of the inputs [x1,x2,x3] ?

function [c,ceq] = constraint_new(X)
  D_i=X(1);D_io=.0014;D_o=X(2);D_oo=.0075;L=X(3);

Looks like the input is being coped into D_* variables,

ceq =[heatload1_new(D_i,D_io,D_o,D_oo,L)-Qreq];

and the D_* variables are being passed into heatload1_new, so the problem must be in heatload1_new

function [Q]=heatload1_new(D_i,D_io,D_o,D_oo,L)

so the D_* variables make it to inside heatload1_new

X=[0.001;0.006;0.4];  
X=[X(1);X(2);X(3)];
D_i=X(1);
D_io=.0014;
D_o=X(2);
D_oo=.0075;
L=X(3);

... and you create a vector of constant X values and use those constants to overwrite the D_* variables. So the D_* variable values that were passed into heatload1_new() become irrelevant and heatload1_new() always returns the same value independent of its inputs. Therefore the ceq constraint always ends up calculating the same thing independent of the inputs to the nonlinear constraint function. And that value is about -3.24, which can never be 0, so it is impossible for the equality constraint to be satisfied.