布谷鸟算法详细讲解

今天我要讲的内容是布谷鸟算法，英文叫做Cuckoo search (CS algorithm)。首先还是同样，介绍一下这个算法的英文含义， Cuckoo是布谷鸟的意思，啥是布谷鸟呢，是一种叫做布谷的鸟，o(∩_∩)o ，这种鸟她妈很懒，自己生蛋自己不养，一般把它的宝宝扔到别的种类鸟的鸟巢去。但是呢，当孵化后，遇到聪明的鸟妈妈，一看就知道不是亲生的，直接就被鸟妈妈给杀了。于是这群布谷鸟宝宝为了保命，它们就模仿别的种类的鸟叫，让智商或者情商极低的鸟妈妈误认为是自己的亲宝宝，这样它就活下来了。 Search指的是搜索，这搜索可不是谷歌一下，你就知道。而是搜索最优值，举个简单的例子，y=(x-0.5)^2+1，它的最小值是1，位置是(0.5,1)，我们要搜索的就是这个位置。

现在我们应该清楚它是干嘛的了吧，它就是为了寻找最小值而产生的一种算法，有些好装X的人会说，你傻X啊，最小值不是-2a/b吗，用你找啊? 说的不错，确实是，但是要是我们的函数变成 y=sin(x^3+x^2)+e^cos(x^3)+log(tan(x)+10，你怎么办吶？你解不了，就算你求导数，但是你知道怎么解导数等于0吗？所以我们就得引入先进的东西来求最小值。

为了使大家容易理解，我还是用y=(x-0.5)^2+1来举例子，例如我们有4个布谷鸟蛋（也就是4个x坐标），鸟妈妈发现不是自己的宝宝的概率是0.25，我们x的取值范围是[0,1]之间，于是我们就可以开始计算了。

目标：求x在[0,1]之内的函数y=(x-0.5)^2+1最小值

（1）初始化x的位置，随机生成4个x坐标，x1=0.4，x2=0.6，x3=0.8，x4=0.3 ——> X=[0.4, 0.6 ,0.8, 0.3]

（2）求出y1~y4，把x1~x4带入函数，求得Y=[1,31, 1.46, 1.69, 1.265]，并选取当前最小值ymin= y4=1.265

（3）开始定出一个y的最大值为Y_global=INF（无穷大），然后与ymin比较，把Y中最小的位置和值保留，例如Y_global=INF>ymin=1.265，所以令Y_global=1.265

（4）记录Y_global的位置，（0.3，1.265）。

（5）按概率0.25，随机地把X中的值过塞子，选出被发现的蛋。例如第二个蛋被发现x2=0.6，那么他就要随机地变换位子，生成一个随机数，例如0.02，然后把x2=x2+0.02=0.62，之后求出y2=1.4794。那么X就变为了X=[0.4, 0.62 ,0.8, 0.3]，Y=[1,31, 1.4794, 1.69, 1.265]。

（6）进行莱维飞行，这名字听起来挺高大上，说白了，就是把X的位置给随机地改变了。怎么变？有一个公式x=x+alpha*L。

L=S*(X-Y_global)*rand3

S=[rand1*sigma/|rand2|]^(1/beta)

sigma=0.6966

beta=1.5

alpha=0.01

rand1~rand3为正态分布的随机数

然后我们把X=[0.4, 0.6 ,0.8, 0.3]中的x1带入公式，首先随机生成rand1=-1.2371，rand2=-2.1935，rand3=-0.3209，接下来带入公式中，获得x1=0.3985

之后同理计算:

x2=0.6172

x3=0.7889 

x4=0.3030

（7）更新矩阵X，X=[0.3985, 0.6172, 0.7889, 0.3030]

（8）计算Y=[1.3092, 1.4766, 1.6751, 1.2661]，并选取当前最小值ymin= y4=1.2661，然后与ymin比较，把Y中最小的位置和值保留，例如Y_global=1.265<ymin=1.2661，所以令Y_global=1.265

（9）返回步骤（5）用更新的X去循环执行，经过多次计算即可获得y的最优值和的最值位置(x,y)

 代码：

% -----------------------------------------------------------------  
% Cuckoo Search (CS) algorithm by Xin-She Yang and Suash Deb      %  
% Programmed by Xin-She Yang at Cambridge University              %  
% Programming dates: Nov 2008 to June 2009                        %  
% Last revised: Dec  2009   (simplified version for demo only)    %  
% -----------------------------------------------------------------  
% Papers -- Citation Details:  
% 1) X.-S. Yang, S. Deb, Cuckoo search via Levy flights,  
% in: Proc. of World Congress on Nature & Biologically Inspired  
% Computing (NaBIC 2009), December 2009, India,  
% IEEE Publications, USA,  pp. 210-214 (2009).  
% http://arxiv.org/PS_cache/arxiv/pdf/1003/1003.1594v1.pdf   
% 2) X.-S. Yang, S. Deb, Engineering optimization by cuckoo search,  
% Int. J. Mathematical Modelling and Numerical Optimisation,   
% Vol. 1, No. 4, 330-343 (2010).   
% http://arxiv.org/PS_cache/arxiv/pdf/1005/1005.2908v2.pdf  
% ----------------------------------------------------------------%  
% This demo program only implements a standard version of         %  
% Cuckoo Search (CS), as the Levy flights and generation of       %  
% new solutions may use slightly different methods.               %  
% The pseudo code was given sequentially (select a cuckoo etc),   %  
% but the implementation here uses Matlab's vector capability,    %  
% which results in neater/better codes and shorter running time.  %   
% This implementation is different and more efficient than the    %  
% the demo code provided in the book by   
%    "Yang X. S., Nature-Inspired Metaheuristic Algoirthms,       %   
%     2nd Edition, Luniver Press, (2010).                 "       %  
% --------------------------------------------------------------- %  
  
% =============================================================== %  
% Notes:                                                          %  
% Different implementations may lead to slightly different        %  
% behavour and/or results, but there is nothing wrong with it,    %  
% as this is the nature of random walks and all metaheuristics.   %  
% -----------------------------------------------------------------  
  
% Additional Note: This version uses a fixed number of generation %  
% (not a given tolerance) because many readers asked me to add    %  
%  or implement this option.                               Thanks.%                            
function [bestnest,fmin]=cuckoo_search_new(n)  
if nargin<1,  
% Number of nests (or different solutions)  
n=25;  
end  
  
% Discovery rate of alien eggs/solutions  
pa=0.25;  
  
%% Change this if you want to get better results  
N_IterTotal=1000;  
%% Simple bounds of the search domain  
% Lower bounds  
nd=15;   
Lb=-5*ones(1,nd);   
% Upper bounds  
Ub=5*ones(1,nd);  
  
% Random initial solutions  
for i=1:n,  
nest(i,:)=Lb+(Ub-Lb).*rand(size(Lb));  
end  
  
% Get the current best  
fitness=10^10*ones(n,1);  
[fmin,bestnest,nest,fitness]=get_best_nest(nest,nest,fitness);  
  
N_iter=0;  
%% Starting iterations  
for iter=1:N_IterTotal,  
    % Generate new solutions (but keep the current best)  
     new_nest=get_cuckoos(nest,bestnest,Lb,Ub);     
     [fnew,best,nest,fitness]=get_best_nest(nest,new_nest,fitness);  
    % Update the counter  
      N_iter=N_iter+n;   
    % Discovery and randomization  
      new_nest=empty_nests(nest,Lb,Ub,pa) ;  
      
    % Evaluate this set of solutions  
      [fnew,best,nest,fitness]=get_best_nest(nest,new_nest,fitness);  
    % Update the counter again  
      N_iter=N_iter+n;  
    % Find the best objective so far    
    if fnew<fmin,  
        fmin=fnew;  
        bestnest=best;  
    end  
end %% End of iterations  
  
%% Post-optimization processing  
%% Display all the nests  
disp(strcat('Total number of iterations=',num2str(N_iter)));  
fmin  
bestnest  
  
%% --------------- All subfunctions are list below ------------------  
%% Get cuckoos by ramdom walk  
function nest=get_cuckoos(nest,best,Lb,Ub)  
% Levy flights  
n=size(nest,1);  
% Levy exponent and coefficient  
% For details, see equation (2.21), Page 16 (chapter 2) of the book  
% X. S. Yang, Nature-Inspired Metaheuristic Algorithms, 2nd Edition, Luniver Press, (2010).  
beta=3/2;  
sigma=(gamma(1+beta)*sin(pi*beta/2)/(gamma((1+beta)/2)*beta*2^((beta-1)/2)))^(1/beta);  
  
for j=1:n,  
    s=nest(j,:);  
    % This is a simple way of implementing Levy flights  
    % For standard random walks, use step=1;  
    %% Levy flights by Mantegna's algorithm  
    u=randn(size(s))*sigma;  
    v=randn(size(s));  
    step=u./abs(v).^(1/beta);  
    
    % In the next equation, the difference factor (s-best) means that   
    % when the solution is the best solution, it remains unchanged.       
    stepsize=0.01*step.*(s-best);  
    % Here the factor 0.01 comes from the fact that L/100 should the typical  
    % step size of walks/flights where L is the typical lenghtscale;   
    % otherwise, Levy flights may become too aggresive/efficient,   
    % which makes new solutions (even) jump out side of the design domain   
    % (and thus wasting evaluations).  
    % Now the actual random walks or flights  
    s=s+stepsize.*randn(size(s));  
   % Apply simple bounds/limits  
   nest(j,:)=simplebounds(s,Lb,Ub);  
end  
  
%% Find the current best nest  
function [fmin,best,nest,fitness]=get_best_nest(nest,newnest,fitness)  
% Evaluating all new solutions  
for j=1:size(nest,1),  
    fnew=fobj(newnest(j,:));  
    if fnew<=fitness(j),  
       fitness(j)=fnew;  
       nest(j,:)=newnest(j,:);  
    end  
end  
% Find the current best  
[fmin,K]=min(fitness) ;  
best=nest(K,:);  
  
%% Replace some nests by constructing new solutions/nests  
function new_nest=empty_nests(nest,Lb,Ub,pa)  
% A fraction of worse nests are discovered with a probability pa  
n=size(nest,1);  
% Discovered or not -- a status vector  
K=rand(size(nest))>pa;  
  
% In the real world, if a cuckoo's egg is very similar to a host's eggs, then   
% this cuckoo's egg is less likely to be discovered, thus the fitness should   
% be related to the difference in solutions.  Therefore, it is a good idea   
% to do a random walk in a biased way with some random step sizes.    
%% New solution by biased/selective random walks  
stepsize=rand*(nest(randperm(n),:)-nest(randperm(n),:));  
new_nest=nest+stepsize.*K;  
for j=1:size(new_nest,1)  
    s=new_nest(j,:);  
  new_nest(j,:)=simplebounds(s,Lb,Ub);    
end  
  
% Application of simple constraints  
function s=simplebounds(s,Lb,Ub)  
  % Apply the lower bound  
  ns_tmp=s;  
  I=ns_tmp<Lb;  
  ns_tmp(I)=Lb(I);  
    
  % Apply the upper bounds   
  J=ns_tmp>Ub;  
  ns_tmp(J)=Ub(J);  
  % Update this new move   
  s=ns_tmp;  
  
%% You can replace the following by your own functions  
% A d-dimensional objective function  
function z=fobj(u)  
%% d-dimensional sphere function sum_j=1^d (u_j-1)^2.   
%  with a minimum at (1,1, ...., 1);   
z=sum((u-1).^2); 

版权声明：本文为博主原创文章，未经博主允许不得转载。 http://blog.csdn.net/u013631121/article/details/76944879