Reference to non-existent field although the field is determined

Question

raha ahmadi 2021-2-12

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/744022-reference-to-non-existent-field-although-the-field-is-determined

评论： raha ahmadi 2021-2-13

Hi, I m working on a POS algorithm for optimizing an elemet of a matrix which is produced of multiplication of almost 1000 matrices.

I think my PSO code is correct but I don’t know why I get this error?

Could it be because of my CostFunction which is resulted from multiplication of 1000 matrices? I determine the filed of the structure but nothing happened?

Reference to non-existent field 'Position'.

Error in PSO (line 95)

+c2*rand(VarSize).*(GlobalBest.Position-particle(i).Position);

Thanks in advance

Raha

2 个评论
显示无隐藏无

G A 2021-2-12

GlobalBest.Position

is not defined

raha ahmadi 2021-2-13

Dear GA

Thank you for your answer but I defined it in line 67:

GlobalBest=particle(i).Best;

which means:

GlobalBest.Position=particle(i).Best.Position;

and

GlobalBest.Cost=particle(i).Best.Cost

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Walter Roberson 2021-2-13

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/744022-reference-to-non-existent-field-although-the-field-is-determined#answer_622442

编辑：Walter Roberson 2021-2-13

在 MATLAB Online 中打开

GlobalBest=particle(i).Best;

You do that conditionally on some particle cost being found that is less than the initialized cost of infinity. However, that never comes true: your calculated costs for your initial population are all nan +/- nan*1i (complex nan).

Nth=1.4786e24;

You are working in ranges about 1e23 to 1e25.

g1=a1*(N-N0)-a2*(lambda-(lambdaB-a3*(N-Nth)))^2;

lambda-(lambdaB-a3*(N-Nth)) is about 750. Squaring that is on the order of 1E5. a2 is on the order of 1e19, so the product of the two is on the order of 1e24

gamma1=g1/2+1i*n1*omega/c;

That is the same order of magnitude, with a small (roughly pi) complex component

M1=[exp(gamma1*pitch/2),0;0,exp(-gamma1*pitch/2)]; % in environment 1

exp(gamma1*pitch/2) and exp(-gamma1*pitch/2) are the same but with different signs on the exponent. If gamma1 is about 1e24 and negative, exp(gamma1*pitch/2) would be roughly exp(-1e24) which would come out as 0, but exp(-gamm1*pitch/2) would be roughly exp(1e24) which would overflow to infinity. If gamma1 is about -1e24, then which of the two exp() is 0 and which is inf switches roles. Either way, one of the two is going to be infinite when you are working in that range. And that is going to poison all the rest of the calculations, leading to NaN results.

You have large values on the order of 1e24 hard-coded, you aren't going to get around those at all easy; no mere typing mistake. But to get finite values out of the expression, your pitch would have to down-scale the 1e24 into the range of no more than 700-ish.

Or, you would have to switch to the symbolic toolbox for that calculation. For the symbolic toolbox, the limit is roughly

>> vpa(exp(sym(7.4e8)))

ans =

4.0588813157826340060476320819064e321377916

which gets you an extra factor of 1e6 to play with.

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

raha ahmadi 2021-2-13

Dear Walter Roberson

I really appreciate the time you spent explaining the details. I don't know how to thank you. I wrongly thought that the problem is calculation of the matrices multiplication.

Wish you all the best

请先登录，再进行评论。