Question

我正在尝试使用Matlab GA（在全局优化工具箱中）解决广义分配问题（将m个代理分配给n个任务）。为了减少计算时间，我创建了一个满足行和列约束的初始总体矩阵（MM），并将GA选项“ InitialPopulationMatrix”设置为MM。

opts = optimoptions(@ga,'InitialPopulationMatrix',MM);

但是，GA解决方案比我的最初人群差很多。首先，GA解的适合度值为7625，而在初始总体中，该值仅为496。其次，GA解决方案无法满足大多数约束条件。另外，当增加最大发电量时，GA解决方案不会变得更好，就像被卡在本地一样。

要获得良好的GA解决方案，如何通过设置一些GA选项来改善编码？

我的问题是大规模的，即m = 60，n = 145，二进制变量的数量是8700（60 * 145），赋值成本是二进制变量的函数（导致二次方）目标函数）。

下面是主要脚本：

%Parameter Setting
tic
nvars=8700;
b=[-dli;dui;-sj;sj];
lb=zeros(1,nvars);
ub=ones(1,nvars);
IntCon=[1:nvars];
%build constraint matrix
  %ga evaluates the matrix product A*x as if x is transposed (A*x').
Ar=kron(eye(nShift),ones(1,nCombo));%Summation along rows of Xij can be expressed as a matrix-vector multiplication A1*XijT(:) 
Ac=kron(ones(1,nShift),eye(nCombo));% Summation along columns can be expressed the same way with

  %Note:  for integer programming, Aeq and beq must be empty ([]), so
  %equality contraints are rewritten to si<=Ac<=sj
A=[-Ar;Ar;-Ac;Ac];
nonlcon = [];
MM=IM(dli,dui,sj);
%PopulationSize: Positive integer | {50} whennumberOfVariables <= 5, {200} otherwise |{min(max(10*nvars,40),100)} for mixed-integer problems

opts = optimoptions(@ga, ...
                    'PopulationSize', 20, ...
                    'MaxGenerations', 10000, ...
                    'InitialPopulationRange',[0;1],...
                    'CrossoverFraction',0.9,...
                    'EliteCount',10,...
                    'InitialPopulationMatrix',MM,...
                    'PlotFcn', {@gaplotbestf,@gaplotstopping});

[x,faval,exitflag]=ga(@Obj_f,nvars,A,b,[],[],lb,ub,nonlcon,IntCon,opts)
toc
Time=toc;

这是我用于健身功能的Obj_f.m

function [TotalV] = Obj_f(x)
%Time halo violation cost of using matrix Xij
nShift=60;
nCombo=145;
t=3;
%Convert the variable vector x to matrices with nCombo columns.
Xij=vec2mat(x,nCombo);
  %Compute cij for the initial xij_Feasible matrix
Cij=[Xij;Xij(1:t-1,:)];%add the remaining t-1 element's cost into matrix Xij
  [m,n]=size(Xij);
  cij=zeros(m,n);

for j=1:nCombo
    for i=1:nShift
        cij(i,j)=sum(Cij((i:i+t-1),j));
    end
end

Violation=cij.*Xij;
TotalV=sum(Violation(:));

end

为什么我的遗传算法解决方案比给定的初始种群差很多？如何设置选项来解决此问题？

0 个答案: