多目标优化算法测试基准函数（CEC2009）以及评价标准

已于 2023-12-06 18:45:29 修改
阅读量3.3k 收藏 41
点赞数 30
分类专栏： 算法 文章标签： 算法
于 2023-11-30 20:35:59 首次发布
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/weixin_47577143/article/details/134713055
版权

前言

        IEEE进化计算会议（IEEE on Evolutionary Computation，IEEE CEC）从2005年开始，后面每年举办一次，其中定义了很多测试案例，被用来作为各种进化算法的测试基准函数，下面将介绍2009年的多目标测试函数（UF1-UF10,以及CF1-CF10）以及相关性能评价标准。

1. 多目标测试集定义

1.1. 双目标测试函数

  • UF1

\left\{ \begin{aligned} &f_{1}(x)=x_{1}+\frac{2}{|J_{1}|}\sum_{j\in J_{1}}[x_{j}-sin(6\pi x_{1}+\frac{j\pi}{n})]^{2} \\ &f_{2}(x)=1-\sqrt x_{1} + \frac{2}{|J_{2}|}\sum_{j\in J_{2}}[x_{j}-sin(6\pi x_{1}+\frac{j\pi}{n})]^{2} \\ &J_{1}=\left\{j|j \,is \,odd \,and \,2 \leq j \leq n \right\}\\ &J_{2}=\left\{j|j \,is \,even \,and \,2 \leq j \leq n \right\}\\ \end{aligned} \right.

  • UF2

\left\{ \begin{aligned} &f_{1}(x)=x_{1}+\frac{2}{|J_{1}|}\sum_{j\in J_{1}}y^{2}_{j}\\ &f_{2}(x)=1-\sqrt x_{1} + \frac{2}{|J_{2}|}\sum_{j\in J_{2}}y^{2}_{j} \\ &J_{1}=\left\{j|j \,is \,odd \,and \,2 \leq j \leq n \right\}\\ &J_{2}=\left\{j|j \,is \,even \,and \,2 \leq j \leq n \right\}\\ \end{aligned} \right.

y_j=\left\{ \begin{aligned} &x_{j}-[0.3x^{2}_{1}cos(24\pi x_{1}+\frac{4j\pi}{n})+0.6x_{1}]cos(6\pi x_{1}+\frac{j\pi}{n})) \, j \in J_{1}\\ &x_{j}-[0.3x^{2}_{1}cos(24\pi x_{1}+\frac{4j\pi}{n})+0.6x_{1}]sin(6\pi x_{1}+\frac{j\pi}{n})) \, j \in J_{2}\\ \end{aligned} \right.

  • UF3

\left\{ \begin{aligned} &f_{1}(x)=x_{1}+\frac{2}{|J_{1}|}(4\sum_{j\in J_{1}}y^{2}_{j}-2 \prod_{j\in J_{1}}cos(\frac{20y_{j} \pi}{\sqrt j})+2) \\ &f_{2}(x)=\sqrt x_{1}+\frac{2}{|J_{2}|}(4\sum_{j\in J_{1}}y^{2}_{j}-2 \prod_{j\in J_{2}}cos(\frac{20y_{j} \pi}{\sqrt j})+2) \\ &J_{1} \, and \, J_{2} \, are \, the \, same \, as \, those \,of \, UF1 \end{aligned} \right.

y_{j}=x_{j}-x^{0.5(1+\frac{3(j-2)}{n-2})}_{1},\,j=2,3,...,n

  • UF4

\left\{ \begin{aligned} &f_{1}(x)=x_{1}+\frac{2}{|J_{1}|}\sum_{j\in J_{1}}h(y_{j})) \\ &f_{2}(x)=1- x_{2} + \frac{2}{|J_{2}|}\sum_{j\in J_{2}}h(y_{j})\\ &J_{1} \, and \, J_{2} \,are \,the \,same \,as \,those \,of \,UF1 \end{aligned} \right.

y_{j}=x_{j}-sin(6\pi x_{1}+\frac{j\pi}{n}),\,j=2,3,...,n,\, h(t)=\frac{|t|}{1+e^{2|t|}}

  • UF5

\left\{ \begin{aligned} &f_{1}(x)=x_{1}+(\frac{1}{2N}+\in)|sin(2N\pi x_{1})|+\frac{2}{|J_{1}|}\sum_{j\in J_{1}}h(y_{j})) \\ &f_{2}(x)=1-x_{1}+(\frac{1}{2N}+\in)|sin(2N\pi x_{1})|+\frac{2}{|J_{1}|}\sum_{j\in J_{2}}h(y_{j})) \\ &y_{j}=x_{j}-sin(6\pi x_{1}+\frac{j\pi}{n}),\,j=2,3,...,n,\,\\ &h(t)=2t^2-cos(4\pi t)+1\\ &J_{1} \, and \, J_{2} \,are \,the \,same \,as \,those \,of \,UF1 ,\in>0 \end{aligned} \right.

  • UF6

\left\{ \begin{aligned} &f_{1}(x)=x_{1}+max\left\{0,2(\frac{1}{2N}+\in)sin(2N\pi x)\right\}+\frac{2}{|J_{1}|}(4\sum_{j\in J_{1}}y^{2}_{j}-2 \prod_{j\in J_{1}}cos(\frac{20y_{j} \pi}{\sqrt j})+2) \\ &f_{2}(x)=1-x_{1}+max\left\{0,2(\frac{1}{2N}+\in)sin(2N\pi x)\right\}+\frac{2}{|J_{1}|}(4\sum_{j\in J_{1}}y^{2}_{j}-2 \prod_{j\in J_{1}}cos(\frac{20y_{j} \pi}{\sqrt j})+2) \\ &y_{j}=x_{j}-sin(6\pi x_{1} + \frac{j\pi}{n}),j=2,3,...,n\\ &J_{1} \, and \, J_{2} \, are \, the \, same \, as \, those \,of \, UF1,\in>0 \end{aligned} \right.

  • UF7

\left\{ \begin{aligned} &f_{1}(x)=\sqrt[5]{x_{1}}+\frac{2}{|J_{1}|}\sum_{j\in J_{1}}y^{2}_{j}\\ &f_{2}(x)=1-\sqrt[5]{ x_{1}} + \frac{2}{|J_{2}|}\sum_{j\in J_{2}}y^{2}_{j} \\ &y_{j}=x_{j}-sin(6\pi x_{1} + \frac{j\pi}{n}),j=2,3,...,n\\ &J_{1} \, and \, J_{2} \, are \, the \, same \, as \, those \,of \, UF1,\in>0 \end{aligned} \right.

1.2. 三目标测试函数

  • UF8

\left\{ \begin{aligned} &f_{1}(x)=cos(0.5x_{1}\pi)cos(0.5x_{2}\pi)+\frac{2}{|J_{1}|}\sum_{j\in J_{1}}(x_{j}-2x_{2}sin(2\pi x_{1}+\frac{j\pi}{n})^{2}) \\ &f_{2}(x)=cos(0.5x_{1}\pi)cos(0.5x_{2}\pi)+\frac{2}{|J_{2}|}\sum_{j\in J_{2}}(x_{j}-2x_{2}sin(2\pi x_{1}+\frac{j\pi}{n})^{2}) \\ &f_{3}(x)=sin(0.5x_{1}\pi)+\frac{2}{|J_{3}|}\sum_{j\in J_{3}}(x_{j}-2x_{2}sin(2\pi x_{1}+\frac{j\pi}{n})^{2}) \\ &J_{1}=\left\{j|3\leq j \leq n,and \,j-1\,is \,a \,multiplication \,of \,3 \right\}\\ &J_{2}=\left\{j|3\leq j \leq n,and \,j-2\,is \,a \,multiplication \,of \,3 \right\}\\ &J_{2}=\left\{j|3\leq j \leq n,and \,j\,is \,a \,multiplication \,of \,3 \right\}\\ \end{aligned} \right.

  • UF9

\left\{ \begin{aligned} &f_{1}(x)=0.5[max\left\{0,(1+\in)(1-4(2x_{1}-1)^{2})\right\}+2x_{1}]x_{2}+\frac{2}{|J_{1}|}\sum_{j\in J_{1}}(x_{j}-2x_{2}sin(2\pi x_{1}+\frac{j\pi}{n})^{2}) \\ &f_{2}(x)=0.5[max\left\{0,(1+\in)(1-4(2x_{1}-1)^{2})\right\}+2x_{1}]x_{2}+\frac{2}{|J_{2}|}\sum_{j\in J_{2}}(x_{j}-2x_{2}sin(2\pi x_{1}+\frac{j\pi}{n})^{2}) \\ &f_{3}(x)=1-x_{2}+\frac{2}{|J_{3}|}\sum_{j\in J_{3}}(x_{j}-2x_{2}sin(2\pi x_{1}+\frac{j\pi}{n})^{2}) \\ &J_{1}=\left\{j|3\leq j \leq n,and \,j-1\,is \,a \,multiplication \,of \,3 \right\}\\ &J_{2}=\left\{j|3\leq j \leq n,and \,j-2\,is \,a \,multiplication \,of \,3 \right\}\\ &J_{2}=\left\{j|3\leq j \leq n,and \,j\,is \,a \,multiplication \,of \,3 \right\},\in=0.1\end{aligned} \right.

  • UF10

\left\{ \begin{aligned} &f_{1}(x)=cos(0.5x_{1}\pi)cos(0.5x_{2}\pi)+\frac{2}{|J_{1}|}\sum_{j\in J_{1}}[4y^{2}_{j}-cos(8\pi y_{j})+1]\\ &f_{2}(x)=cos(0.5x_{1}\pi)cos(0.5x_{2}\pi)+\frac{2}{|J_{2}|}\sum_{j\in J_{2}}[4y^{2}_{j}-cos(8\pi y_{j})+1]\\ &f_{3}(x)=sin(0.5x_{1}\pi)+\frac{2}{|J_{3}|}\sum_{j\in J_{3}}[4y^{2}_{j}-cos(8\pi y_{j})+1] \\ &J_{1}=\left\{j|3\leq j \leq n,and \,j-1\,is \,a \,multiplication \,of \,3 \right\}\\ &J_{2}=\left\{j|3\leq j \leq n,and \,j-2\,is \,a \,multiplication \,of \,3 \right\}\\ &J_{2}=\left\{j|3\leq j \leq n,and \,j\,is \,a \,multiplication \,of \,3 \right\}\\ \end{aligned} \right.

带约束的问题，即CF1-CF10，这里就不累述，感兴趣的， 可以下载附件研读。

2. 评价标准

        在评价算法的性能时，最常见的通过下面几个标准进行定义：

  • Generational Distance(GD)

        

  • Inverted Generational Distance(IGD)

  • Spacing (SP)  

3. 测试函数code

3.1.CEC2009

% cec09.m
% 
% The Matlab version of the test instances for CEC 2009 Multiobjective
%   Optimization Competition.
% 
% Usage: fobj = cec09(problem_name), the handle of the function will be
%   with fobj
% 
% Please refer to the report for correct one if the source codes are not
%   consist with the report.
% History:
%   v1 Sept.08 2008
%   v2 Nov.18  2008
%   v3 Nov.26  2008

function fobj = cec09(name)

    switch name
        case 'UF1'
            fobj = @UF1;
        case 'UF2'
            fobj = @UF2; 
        case 'UF3'
            fobj = @UF3;  
        case 'UF4'
            fobj = @UF4;
        case 'UF5'
            fobj = @UF5; 
        case 'UF6'
            fobj = @UF6;
        case 'UF7'
            fobj = @UF7;
        case 'UF8'
            fobj = @UF8; 
        case 'UF9'
            fobj = @UF9; 
        case 'UF10'
            fobj = @UF10;
        case 'CF1'
            fobj = @CF1;
        case 'CF2'
            fobj = @CF2; 
        case 'CF3'
            fobj = @CF3;  
        case 'CF4'
            fobj = @CF4;
        case 'CF5'
            fobj = @CF5; 
        case 'CF6'
            fobj = @CF6;
        case 'CF7'
            fobj = @CF7;
        case 'CF8'
            fobj = @CF8; 
        case 'CF9'
            fobj = @CF9; 
        case 'CF10'
            fobj = @CF10;           
        otherwise
            fobj = @UF1;
    end
end

%% UF1
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function y = UF1(x)
    [dim, num]  = size(x);
    tmp         = zeros(dim,num);
    tmp(2:dim,:)= (x(2:dim,:) - sin(6.0*pi*repmat(x(1,:),[dim-1,1]) + pi/dim*repmat((2:dim)',[1,num]))).^2;
    tmp1        = sum(tmp(3:2:dim,:));  % odd index
    tmp2        = sum(tmp(2:2:dim,:));  % even index
    y(1,:)      = x(1,:)             + 2.0*tmp1/size(3:2:dim,2);
    y(2,:)      = 1.0 - sqrt(x(1,:)) + 2.0*tmp2/size(2:2:dim,2);
    clear tmp;
end

%% UF2
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function y = UF2(x)
    [dim, num]  = size(x);
    X1          = repmat(x(1,:),[dim-1,1]);
    A           = 6*pi*X1 + pi/dim*repmat((2:dim)',[1,num]);
    tmp         = zeros(dim,num);    
    tmp(2:dim,:)= (x(2:dim,:) - 0.3*X1.*(X1.*cos(4.0*A)+2.0).*cos(A)).^2;
    tmp1        = sum(tmp(3:2:dim,:));  % odd index
    tmp(2:dim,:)= (x(2:dim,:) - 0.3*X1.*(X1.*cos(4.0*A)+2.0).*sin(A)).^2;
    tmp2        = sum(tmp(2:2:dim,:));  % even index
    y(1,:)      = x(1,:)             + 2.0*tmp1/size(3:2:dim,2);
    y(2,:)      = 1.0 - sqrt(x(1,:)) + 2.0*tmp2/size(2:2:dim,2);
    clear X1 A tmp;
end

%% UF3
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function y = UF3(x)
    [dim, num]   = size(x);
    Y            = zeros(dim,num);
    Y(2:dim,:)   = x(2:dim,:) - repmat(x(1,:),[dim-1,1]).^(0.5+1.5*(repmat((2:dim)',[1,num])-2.0)/(dim-2.0));
    tmp1         = zeros(dim,num);
    tmp1(2:dim,:)= Y(2:dim,:).^2;
    tmp2         = zeros(dim,num);
    tmp2(2:dim,:)= cos(20.0*pi*Y(2:dim,:)./sqrt(repmat((2:dim)',[1,num])));
    tmp11        = 4.0*sum(tmp1(3:2:dim,:)) - 2.0*prod(tmp2(3:2:dim,:)) + 2.0;  % odd index
    tmp21        = 4.0*sum(tmp1(2:2:dim,:)) - 2.0*prod(tmp2(2:2:dim,:)) + 2.0;  % even index
    y(1,:)       = x(1,:)             + 2.0*tmp11/size(3:2:dim,2);
    y(2,:)       = 1.0 - sqrt(x(1,:)) + 2.0*tmp21/size(2:2:dim,2);
    clear Y tmp1 tmp2;
end

%% UF4
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function y = UF4(x)
    [dim, num]  = size(x);
    Y           = zeros(dim,num);
    Y(2:dim,:)  = x(2:dim,:) - sin(6.0*pi*repmat(x(1,:),[dim-1,1]) + pi/dim*repmat((2:dim)',[1,num]));
    H           = zeros(dim,num);
    H(2:dim,:)  = abs(Y(2:dim,:))./(1.0+exp(2.0*abs(Y(2:dim,:))));
    tmp1        = sum(H(3:2:dim,:));  % odd index
    tmp2        = sum(H(2:2:dim,:));  % even index
    y(1,:)      = x(1,:)          + 2.0*tmp1/size(3:2:dim,2);
    y(2,:)      = 1.0 - x(1,:).^2 + 2.0*tmp2/size(2:2:dim,2);
    clear Y H;
end

%% UF5
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function y = UF5(x)
    N           = 10.0;
    E           = 0.1;
    [dim, num]  = size(x);
    Y           = zeros(dim,num);
    Y(2:dim,:)  = x(2:dim,:) - sin(6.0*pi*repmat(x(1,:),[dim-1,1]) + pi/dim*repmat((2:dim)',[1,num]));
    H           = zeros(dim,num);
    H(2:dim,:)  = 2.0*Y(2:dim,:).^2 - cos(4.0*pi*Y(2:dim,:)) + 1.0;
    tmp1        = sum(H(3:2:dim,:));  % odd index
    tmp2        = sum(H(2:2:dim,:));  % even index
    tmp         = (0.5/N+E)*abs(sin(2.0*N*pi*x(1,:)));
    y(1,:)      = x(1,:)      + tmp + 2.0*tmp1/size(3:2:dim,2);
    y(2,:)      = 1.0 - x(1,:)+ tmp + 2.0*tmp2/size(2:2:dim,2);
    clear Y H;
end

%% UF6
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function y = UF6(x)
    N            = 2.0;
    E            = 0.1;
    [dim, num]   = size(x);
    Y            = zeros(dim,num);
    Y(2:dim,:)  = x(2:dim,:) - sin(6.0*pi*repmat(x(1,:),[dim-1,1]) + pi/dim*repmat((2:dim)',[1,num]));
    tmp1         = zeros(dim,num);
    tmp1(2:dim,:)= Y(2:dim,:).^2;
    tmp2         = zeros(dim,num);
    tmp2(2:dim,:)= cos(20.0*pi*Y(2:dim,:)./sqrt(repmat((2:dim)',[1,num])));
    tmp11        = 4.0*sum(tmp1(3:2:dim,:)) - 2.0*prod(tmp2(3:2:dim,:)) + 2.0;  % odd index
    tmp21        = 4.0*sum(tmp1(2:2:dim,:)) - 2.0*prod(tmp2(2:2:dim,:)) + 2.0;  % even index
    tmp          = max(0,(1.0/N+2.0*E)*sin(2.0*N*pi*x(1,:)));
    y(1,:)       = x(1,:)       + tmp + 2.0*tmp11/size(3:2:dim,2);
    y(2,:)       = 1.0 - x(1,:) + tmp + 2.0*tmp21/size(2:2:dim,2);
    clear Y tmp1 tmp2;
end

%% UF7
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function y = UF7(x)
    [dim, num]  = size(x);
    Y           = zeros(dim,num);
    Y(2:dim,:)  = (x(2:dim,:) - sin(6.0*pi*repmat(x(1,:),[dim-1,1]) + pi/dim*repmat((2:dim)',[1,num]))).^2;
    tmp1        = sum(Y(3:2:dim,:));  % odd index
    tmp2        = sum(Y(2:2:dim,:));  % even index
    tmp         = (x(1,:)).^0.2;
    y(1,:)      = tmp       + 2.0*tmp1/size(3:2:dim,2);
    y(2,:)      = 1.0 - tmp + 2.0*tmp2/size(2:2:dim,2);
    clear Y;
end

%% UF8
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function y = UF8(x)
    [dim, num]  = size(x);
    Y           = zeros(dim,num);
    Y(3:dim,:)  = (x(3:dim,:) - 2.0*repmat(x(2,:),[dim-2,1]).*sin(2.0*pi*repmat(x(1,:),[dim-2,1]) + pi/dim*repmat((3:dim)',[1,num]))).^2;
    tmp1        = sum(Y(4:3:dim,:));  % j-1 = 3*k
    tmp2        = sum(Y(5:3:dim,:));  % j-2 = 3*k
    tmp3        = sum(Y(3:3:dim,:));  % j-0 = 3*k
    y(1,:)      = cos(0.5*pi*x(1,:)).*cos(0.5*pi*x(2,:)) + 2.0*tmp1/size(4:3:dim,2);
    y(2,:)      = cos(0.5*pi*x(1,:)).*sin(0.5*pi*x(2,:)) + 2.0*tmp2/size(5:3:dim,2);
    y(3,:)      = sin(0.5*pi*x(1,:))                     + 2.0*tmp3/size(3:3:dim,2);
    clear Y;
end

%% UF9
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function y = UF9(x)
    E           = 0.1;
    [dim, num]  = size(x);
    Y           = zeros(dim,num);
    Y(3:dim,:)  = (x(3:dim,:) - 2.0*repmat(x(2,:),[dim-2,1]).*sin(2.0*pi*repmat(x(1,:),[dim-2,1]) + pi/dim*repmat((3:dim)',[1,num]))).^2;
    tmp1        = sum(Y(4:3:dim,:));  % j-1 = 3*k
    tmp2        = sum(Y(5:3:dim,:));  % j-2 = 3*k
    tmp3        = sum(Y(3:3:dim,:));  % j-0 = 3*k
    tmp         = max(0,(1.0+E)*(1-4.0*(2.0*x(1,:)-1).^2));
    y(1,:)      = 0.5*(tmp+2*x(1,:)).*x(2,:)     + 2.0*tmp1/size(4:3:dim,2);
    y(2,:)      = 0.5*(tmp-2*x(1,:)+2.0).*x(2,:) + 2.0*tmp2/size(5:3:dim,2);
    y(3,:)      = 1-x(2,:)                       + 2.0*tmp3/size(3:3:dim,2);
    clear Y;
end

%% UF10
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function y = UF10(x)
    [dim, num]  = size(x);
    Y           = zeros(dim,num);
    Y(3:dim,:)  = x(3:dim,:) - 2.0*repmat(x(2,:),[dim-2,1]).*sin(2.0*pi*repmat(x(1,:),[dim-2,1]) + pi/dim*repmat((3:dim)',[1,num]));
    H           = zeros(dim,num);
    H(3:dim,:)  = 4.0*Y(3:dim,:).^2 - cos(8.0*pi*Y(3:dim,:)) + 1.0;
    tmp1        = sum(H(4:3:dim,:));  % j-1 = 3*k
    tmp2        = sum(H(5:3:dim,:));  % j-2 = 3*k
    tmp3        = sum(H(3:3:dim,:));  % j-0 = 3*k
    y(1,:)      = cos(0.5*pi*x(1,:)).*cos(0.5*pi*x(2,:)) + 2.0*tmp1/size(4:3:dim,2);
    y(2,:)      = cos(0.5*pi*x(1,:)).*sin(0.5*pi*x(2,:)) + 2.0*tmp2/size(5:3:dim,2);
    y(3,:)      = sin(0.5*pi*x(1,:))                     + 2.0*tmp3/size(3:3:dim,2);
    clear Y H;
end

%% CF1
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function [y,c] = CF1(x)
    a            = 1.0;
    N            = 10.0;
    [dim, num]   = size(x);
    Y            = zeros(dim,num);
    Y(2:dim,:)   = (x(2:dim,:) - repmat(x(1,:),[dim-1,1]).^(0.5+1.5*(repmat((2:dim)',[1,num])-2.0)/(dim-2.0))).^2;
    tmp1         = sum(Y(3:2:dim,:));% odd index
    tmp2         = sum(Y(2:2:dim,:));% even index 
    y(1,:)       = x(1,:)       + 2.0*tmp1/size(3:2:dim,2);
    y(2,:)       = 1.0 - x(1,:) + 2.0*tmp2/size(2:2:dim,2);
    c(1,:)       = y(1,:) + y(2,:) - a*abs(sin(N*pi*(y(1,:)-y(2,:)+1.0))) - 1.0;
    clear Y;
end

%% CF2
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function [y,c] = CF2(x)
    a           = 1.0;
    N           = 2.0;
    [dim, num]  = size(x);
    tmp         = zeros(dim,num);
    tmp(2:dim,:)= (x(2:dim,:) - sin(6.0*pi*repmat(x(1,:),[dim-1,1]) + pi/dim*repmat((2:dim)',[1,num]))).^2;
    tmp1        = sum(tmp(3:2:dim,:));  % odd index
    tmp(2:dim,:)= (x(2:dim,:) - cos(6.0*pi*repmat(x(1,:),[dim-1,1]) + pi/dim*repmat((2:dim)',[1,num]))).^2;
    tmp2        = sum(tmp(2:2:dim,:));  % even index
    y(1,:)      = x(1,:)             + 2.0*tmp1/size(3:2:dim,2);
    y(2,:)      = 1.0 - sqrt(x(1,:)) + 2.0*tmp2/size(2:2:dim,2);
    t           = y(2,:) + sqrt(y(1,:)) - a*sin(N*pi*(sqrt(y(1,:))-y(2,:)+1.0)) - 1.0;
    c(1,:)      = sign(t).*abs(t)./(1.0+exp(4.0*abs(t)));
    clear tmp;
end

%% CF3
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function [y,c] = CF3(x)
    a            = 1.0;
    N            = 2.0;
    [dim, num]   = size(x);
    Y            = zeros(dim,num);
    Y(2:dim,:)   = x(2:dim,:) - sin(6.0*pi*repmat(x(1,:),[dim-1,1]) + pi/dim*repmat((2:dim)',[1,num]));
    tmp1         = zeros(dim,num);
    tmp1(2:dim,:)= Y(2:dim,:).^2;
    tmp2         = zeros(dim,num);
    tmp2(2:dim,:)= cos(20.0*pi*Y(2:dim,:)./sqrt(repmat((2:dim)',[1,num])));
    tmp11        = 4.0*sum(tmp1(3:2:dim,:)) - 2.0*prod(tmp2(3:2:dim,:)) + 2.0;  % odd index
    tmp21        = 4.0*sum(tmp1(2:2:dim,:)) - 2.0*prod(tmp2(2:2:dim,:)) + 2.0;  % even index
    y(1,:)       = x(1,:)          + 2.0*tmp11/size(3:2:dim,2);
    y(2,:)       = 1.0 - x(1,:).^2 + 2.0*tmp21/size(2:2:dim,2);
    c(1,:)       = y(2,:) + y(1,:).^2 - a*sin(N*pi*(y(1,:).^2-y(2,:)+1.0)) - 1.0;   
    clear Y tmp1 tmp2;
end

%% CF4
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function [y,c] = CF4(x)
    [dim, num]  = size(x);
    tmp         = zeros(dim,num);
    tmp(2:dim,:)= x(2:dim,:) - sin(6.0*pi*repmat(x(1,:),[dim-1,1]) + pi/dim*repmat((2:dim)',[1,num]));
    tmp1        = sum(tmp(3:2:dim,:).^2);  % odd index
    tmp2        = sum(tmp(4:2:dim,:).^2);  % even index
    index1      = tmp(2,:) < (1.5-0.75*sqrt(2.0));
    index2      = tmp(2,:)>= (1.5-0.75*sqrt(2.0));
    tmp(2,index1) = abs(tmp(2,index1));
    tmp(2,index2) = 0.125 + (tmp(2,index2)-1.0).^2;
    y(1,:)      = x(1,:)                  + tmp1;
    y(2,:)      = 1.0 - x(1,:) + tmp(2,:) + tmp2;
    t           = x(2,:) - sin(6.0*pi*x(1,:)+2.0*pi/dim) - 0.5*x(1,:) + 0.25;
    c(1,:)      = sign(t).*abs(t)./(1.0+exp(4.0*abs(t)));
    clear tmp index1 index2;
end

%% CF5
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function [y,c] = CF5(x)
    [dim, num]  = size(x);
    tmp         = zeros(dim,num);
    tmp(2:dim,:)= x(2:dim,:) - 0.8*repmat(x(1,:),[dim-1,1]).*cos(6.0*pi*repmat(x(1,:),[dim-1,1]) + pi/dim*repmat((2:dim)',[1,num]));
    tmp1        = sum(2.0*tmp(3:2:dim,:).^2-cos(4.0*pi*tmp(3:2:dim,:))+1.0);  % odd index
    tmp(2:dim,:)= x(2:dim,:) - 0.8*repmat(x(1,:),[dim-1,1]).*sin(6.0*pi*repmat(x(1,:),[dim-1,1]) + pi/dim*repmat((2:dim)',[1,num]));    
    tmp2        = sum(2.0*tmp(4:2:dim,:).^2-cos(4.0*pi*tmp(4:2:dim,:))+1.0);  % even index
    index1      = tmp(2,:) < (1.5-0.75*sqrt(2.0));
    index2      = tmp(2,:)>= (1.5-0.75*sqrt(2.0));
    tmp(2,index1) = abs(tmp(2,index1));
    tmp(2,index2) = 0.125 + (tmp(2,index2)-1.0).^2;
    y(1,:)      = x(1,:)                  + tmp1;
    y(2,:)      = 1.0 - x(1,:) + tmp(2,:) + tmp2;
    c(1,:)      = x(2,:) - 0.8*x(1,:).*sin(6.0*pi*x(1,:)+2.0*pi/dim) - 0.5*x(1,:) + 0.25;
    clear tmp;
end

%% CF6
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function [y,c] = CF6(x)
    [dim, num]  = size(x);
    tmp         = zeros(dim,num);
    tmp(2:dim,:)= x(2:dim,:) - 0.8*repmat(x(1,:),[dim-1,1]).*cos(6.0*pi*repmat(x(1,:),[dim-1,1]) + pi/dim*repmat((2:dim)',[1,num]));
    tmp1        = sum(tmp(3:2:dim,:).^2);  % odd index
    tmp(2:dim,:)= x(2:dim,:) - 0.8*repmat(x(1,:),[dim-1,1]).*sin(6.0*pi*repmat(x(1,:),[dim-1,1]) + pi/dim*repmat((2:dim)',[1,num]));    
    tmp2        = sum(tmp(2:2:dim,:).^2);  % even index
    y(1,:)      = x(1,:)            + tmp1;
    y(2,:)      = (1.0 - x(1,:)).^2 + tmp2;
    tmp         = 0.5*(1-x(1,:))-(1-x(1,:)).^2;
    c(1,:)      = x(2,:) - 0.8*x(1,:).*sin(6.0*pi*x(1,:)+2*pi/dim) - sign(tmp).*sqrt(abs(tmp));
    tmp         = 0.25*sqrt(1-x(1,:))-0.5*(1-x(1,:));
    c(2,:)      = x(4,:) - 0.8*x(1,:).*sin(6.0*pi*x(1,:)+4*pi/dim) - sign(tmp).*sqrt(abs(tmp));    
    clear tmp;
end

%% CF7
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function [y,c] = CF7(x)
    [dim, num]  = size(x);
    tmp         = zeros(dim,num);
    tmp(2:dim,:)= x(2:dim,:) - cos(6.0*pi*repmat(x(1,:),[dim-1,1]) + pi/dim*repmat((2:dim)',[1,num]));
    tmp1        = sum(2.0*tmp(3:2:dim,:).^2-cos(4.0*pi*tmp(3:2:dim,:))+1.0);  % odd index
    tmp(2:dim,:)= x(2:dim,:) - sin(6.0*pi*repmat(x(1,:),[dim-1,1]) + pi/dim*repmat((2:dim)',[1,num]));
    tmp2        = sum(2.0*tmp(6:2:dim,:).^2-cos(4.0*pi*tmp(6:2:dim,:))+1.0);  % even index
    tmp(2,:)    = tmp(2,:).^2;
    tmp(4,:)    = tmp(4,:).^2;
    y(1,:)      = x(1,:)                                  + tmp1;
    y(2,:)      = (1.0 - x(1,:)).^2 + tmp(2,:) + tmp(4,:) + tmp2;
    tmp         = 0.5*(1-x(1,:))-(1-x(1,:)).^2;
    c(1,:)      = x(2,:) - sin(6.0*pi*x(1,:)+2*pi/dim) - sign(tmp).*sqrt(abs(tmp));
    tmp         = 0.25*sqrt(1-x(1,:))-0.5*(1-x(1,:));
    c(2,:)      = x(4,:) - sin(6.0*pi*x(1,:)+4*pi/dim) - sign(tmp).*sqrt(abs(tmp));    
    clear tmp;
end

%% CF8
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function [y,c] = CF8(x)
    N           = 2.0;
    a           = 4.0;
    [dim, num]  = size(x);
    Y           = zeros(dim,num);
    Y(3:dim,:)  = (x(3:dim,:) - 2.0*repmat(x(2,:),[dim-2,1]).*sin(2.0*pi*repmat(x(1,:),[dim-2,1]) + pi/dim*repmat((3:dim)',[1,num]))).^2;
    tmp1        = sum(Y(4:3:dim,:));  % j-1 = 3*k
    tmp2        = sum(Y(5:3:dim,:));  % j-2 = 3*k
    tmp3        = sum(Y(3:3:dim,:));  % j-0 = 3*k
    y(1,:)      = cos(0.5*pi*x(1,:)).*cos(0.5*pi*x(2,:)) + 2.0*tmp1/size(4:3:dim,2);
    y(2,:)      = cos(0.5*pi*x(1,:)).*sin(0.5*pi*x(2,:)) + 2.0*tmp2/size(5:3:dim,2);
    y(3,:)      = sin(0.5*pi*x(1,:))                     + 2.0*tmp3/size(3:3:dim,2);
    c(1,:)      = (y(1,:).^2+y(2,:).^2)./(1.0-y(3,:).^2) - a*abs(sin(N*pi*((y(1,:).^2-y(2,:).^2)./(1.0-y(3,:).^2)+1.0))) - 1.0;
    clear Y;
end

%% CF9
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function [y,c] = CF9(x)
    N           = 2.0;
    a           = 3.0;
    [dim, num]  = size(x);
    Y           = zeros(dim,num);
    Y(3:dim,:)  = (x(3:dim,:) - 2.0*repmat(x(2,:),[dim-2,1]).*sin(2.0*pi*repmat(x(1,:),[dim-2,1]) + pi/dim*repmat((3:dim)',[1,num]))).^2;
    tmp1        = sum(Y(4:3:dim,:));  % j-1 = 3*k
    tmp2        = sum(Y(5:3:dim,:));  % j-2 = 3*k
    tmp3        = sum(Y(3:3:dim,:));  % j-0 = 3*k
    y(1,:)      = cos(0.5*pi*x(1,:)).*cos(0.5*pi*x(2,:)) + 2.0*tmp1/size(4:3:dim,2);
    y(2,:)      = cos(0.5*pi*x(1,:)).*sin(0.5*pi*x(2,:)) + 2.0*tmp2/size(5:3:dim,2);
    y(3,:)      = sin(0.5*pi*x(1,:))                     + 2.0*tmp3/size(3:3:dim,2);
    c(1,:)      = (y(1,:).^2+y(2,:).^2)./(1.0-y(3,:).^2) - a*sin(N*pi*((y(1,:).^2-y(2,:).^2)./(1.0-y(3,:).^2)+1.0)) - 1.0;
    clear Y;
end

%% CF10
% x and y are columnwise, the imput x must be inside the search space and
% it could be a matrix
function [y,c] = CF10(x)
    a           = 1.0;
    N           = 2.0;
    [dim, num]  = size(x);
    Y           = zeros(dim,num);
    Y(3:dim,:)  = x(3:dim,:) - 2.0*repmat(x(2,:),[dim-2,1]).*sin(2.0*pi*repmat(x(1,:),[dim-2,1]) + pi/dim*repmat((3:dim)',[1,num]));
    H           = zeros(dim,num);
    H(3:dim,:)  = 4.0*Y(3:dim,:).^2 - cos(8.0*pi*Y(3:dim,:)) + 1.0;
    tmp1        = sum(H(4:3:dim,:));  % j-1 = 3*k
    tmp2        = sum(H(5:3:dim,:));  % j-2 = 3*k
    tmp3        = sum(H(3:3:dim,:));  % j-0 = 3*k
    y(1,:)      = cos(0.5*pi*x(1,:)).*cos(0.5*pi*x(2,:)) + 2.0*tmp1/size(4:3:dim,2);
    y(2,:)      = cos(0.5*pi*x(1,:)).*sin(0.5*pi*x(2,:)) + 2.0*tmp2/size(5:3:dim,2);
    y(3,:)      = sin(0.5*pi*x(1,:))                     + 2.0*tmp3/size(3:3:dim,2);
    c(1,:)      = (y(1,:).^2+y(2,:).^2)./(1.0-y(3,:).^2) - a*sin(N*pi*((y(1,:).^2-y(2,:).^2)./(1.0-y(3,:).^2)+1.0)) - 1.0;
    clear Y H;
end

3.2 CEC2009 真实Pareto

        在计算评价标准的时候，我们需要用到真实Pareto的解集，Pareto解集一般由以下三种方式获得：

  • 采用穷举法，穷尽最优点
  • 采用多种优化方法进行优化，综合选择最优的点作为Pareto解集
  • 对于一个问题有明确的Pareto解集

        本次的cec2009真实pareto数据集我已经替大家找到，有需要的可以附件直接下载。

总结

       在打UF问题的时候，本人已经尽力不打错，但是难免可能有些小错误，建议还是看一下附件问题描述。本文只是自己的见解以及笔记，有错误的地方，欢迎大家指出。

参考文献：Multi-objective grey wolf optimizer: A novel algorithm for multi-criterion optimization - ScienceDirect

C厂工沛
关注
  • 30
    点赞
  • 踩
  • 41
    收藏
    觉得还不错? 一键收藏
  • 3
    评论
专栏目录
Matlab：多目标优化测试集CEC2009（UF）
weixin_51631044的博客
05-25 1154
MOEA/D 包括分解策略和局部搜索两个部分，通过对目标函数进行分解，将多目标问题转化为许多子问题的优化，然后使用进化算法和局部搜索算法对这些子问题进行处理。它是基于环境选择策略的多目标优化算法，该算法使用拥挤度算子实现多样性维护，有效地处理了多目标问题和种群的多样性问题。当我们面临多个目标时，经常出现在每个目标之间存在牺牲的情况，也就是说，一个目标的优化可能会损害其他目标的优化。为了检验这些算法的性能，就需要标准的测试集来检验，这里提供一个标准的多目标测试集。多目标优化技术可以用于各种领域，比如。
多目标优化算法_多目标优化算法 — 评估指标及测试函数集
weixin_39724009的博客
12-10 3406
点击上方“蓝字”关注我们了解更多精彩评估指标及测试函数集1.相关知识调用评估函数代码如下：%% 评估解：输入(目标函数值，真实前沿值)，输出评估值Scorefunction Score = evaluation_solution(PopObj,PF)global V M pop_sizeE = input('请输入你想评估的指标数字，\n 如：1-PD,2-HV,3-IGD,4-GD...
2024最新多目标优化算法：多目标螳螂搜索算法MOMSA求解46个多目标测试函数+1个工程应用+4种评价指标（提供MATLAB代码）
03-09 1695
它通过计算算法生成的解集中每个解与真实前沿之间的最小距离，并对所有解的距离进行平均来评估算法的性能。与IGD类似，GD也是通过计算算法生成的解集中每个解与真实前沿之间的距离，并对所有解的距离进行平均来评估算法的性能。GD值越小，表示算法生成的解集越接近真实前沿。3. Hypervolume (HV)：HV是一种衡量算法生成的解集所占据的超体积大小的指标。HV值越大，表示算法生成的解集所占据的超体积越大，即解集的多样性和覆盖面积越好。SP值越大，表示算法生成的解集的分布越均匀，解集的多样性和覆盖面积越好。
【最新测试函数】智能优化算法的CEC2022测试函数
最新发布
yuchunyu12的博客
03-22 1538
智能优化算法测试函数是为了在优化算法研究和开发中测试算法性能的规范问题集合。这些测试函数模拟了真实世界优化问题的不同方面，包括局部最小值、最大值、全局最优解，以及多种复杂性如高维度、非线性、不连续等。优化算法，如遗传算法、粒子群优化、蚁群算法和模拟退火等，常常运用这些测试函数来评估其性能。测试函数大致分为几类：单模函数、多模函数、组合函数和约束优化问题。这些测试函数不但在复杂性上有所不同，其形状也各异，有些呈现为平滑曲面，有些则充满尖峰和陡峭的谷地。
算法测试基准函数
10-07
Totally 56 benchmark functions for testing, they are used for intelligence algorithm testing.
多目标优化问题经典测试函数
11-13
多目标优化问题经典测试函数，多个多目标优化问题经典测试函数，用于算法测试，matlab程序
CEC2015：动态多目标测试函数之FDA4、FDA5、FDA5_iso、FDA5_dec
IT猿手
09-19 1239
现实世界中，许多优化问题不仅具有多属性，而且与时间相关，即随着时间的变化，优化问题本身也发生改变，这类问题称为动态多目标优化问题（dynamic multi-objective optimization problems,DMOP)。DMOP是一类目标之间相互冲突，目标函数、约束函数和相关参数等可能随着时间的变化而改变的多目标优化问题。
【多目标进化优化】 MOEA 测试函数
青年有志
12-03 4393
【多目标进化优化】 MOEA 测试函数
粒子群优化算法优化CEC基准测试函数
04-21
CEC基准测试函数是一组用于评价演化计算和群智能算法性能的标准测试集。这些函数具有不同的特性，如多模态、非线性、不连续性和高维性，旨在模拟实际问题的复杂性。CEC测试函数通常分为几类，包括单峰函数、多峰函数...
人工蜂群算法优化CEC基准测试函数
04-21
CEC基准测试函数是一系列设计用于评价进化计算和群体智能算法性能的优化问题集合。这些函数具有不同的特性，如多模态、非线性、不连续、高维度等，能够全面地检验优化算法的搜索能力和稳定性。CEC每年都会发布新的...
多目标测试函数.doc
07-23
我找了好久才找到的，可以说很全了。绝对好用
WFG 多目标测试函数测试集
11-16
WFG多目标测试函数测试集，用于测试多目标性能，
MATLAB:多目标标准测试集CEC2009（UF）
05-25
常用的多目标优化算法有很多种，下面我给你介绍三种当前比较流行的算法。 1. NSGA-II 算法：它是一个经典的多目标优化算法, 需要定义优化目标。NSGA-II 通过对选择和遗传算子进行改进优化，可以在解决多目标优化问题时在时间和空间效率上取得良好的效果。 2. MOEA/D 算法：MOEA/D 包括分解策略和局部搜索两个部分，通过对目标函数进行分解，将多目标问题转化为许多子问题的优化，然后使用进化算法和局部搜索算法对这些子问题进行处理。这种算法可以降低计算复杂度，提高搜索效率。 3. SPEA2 算法：它是基于环境选择策略的多目标优化算法，该算法使用拥挤度算子实现多样性维护，有效地处理了多目标问题和种群的多样性问题。与 NSGA-II 算法相比，其在保持较高解集质量的同时，拥有较高的搜索效率。 为了检验这些算法的性能，就需要标准的测试集来检验，这里提供一个标准的多目标测试集CEC2009（UF）的matlab编写
CEC基准函数——用于算法的测试
03-25
CEC2014；CEC2015；CEC2017；CEC2019；CEC2020，可用于智能优化算法的性能测试
ZDT,DTLZ,MOP，SCH,NOL多目标测函数试集
03-18
1.以m文件的形式编写好的，直接就可以调用
适合解决大规模多目标优化问题的相关算法和相关的基准测试函数、进化多目标算法PlatEMO3.0
11-23
本文将详细介绍用于解决大规模多目标优化问题的相关算法、基准测试函数以及进化多目标优化平台PlatEMO3.0的使用方法。在多目标优化领域，通常需要平衡多个相互冲突的目标，使得每个目标尽可能最优。这涉及到复杂的...
24个基准测试函数matlab
07-08
本资源包含24个这样的函数，它们是优化算法测试的标准集合。 1. **函数的多样性**：这24个函数覆盖了多种类型的优化问题，包括单峰、多峰、有界和无界问题，旨在模拟实际应用中可能出现的各种挑战。例如，一些函数...
CEC2013-benchmark基准测试函数及代码
热门推荐
FemaleHacker的博客
07-20 1万+
可以用CEC13BenchMark来测试自己的算法。 20个具有不同特征的基准测试函数（包括几个相同但尺寸大小不同的函数）来评估小生境算法 函数 F1:F_1:F1​:Five-Uneven-Peak Trap (1D) F2:F_2:F2​:Equal Maxima (1D) F3:F_3:F3​:Uneven Decreasing Maxima (1D) F4:F_4:F4​:Himmelblau (2D) F5:F_5:F5​:Six-Hump Camel Back (2D) F6:F_6:F6​:
28个CEC基本测试函数
01-23
28个CEC基本测试函数，用于测试算法性能，帮助研究算法的同学们
评论 3
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值