ga_approx.m

人工免疫算法基于遗传MATLAB代码很有用哦

% tune the GA in this case by changing the 1 in the numerator to another 
% constant and the .1 to a different value.

sumfitness = sumfitness + fitness(chrom_number);  % Store this for 
      											  % use below
   end

% Next, determine the most fit and least fit chromosome and 
% the chrom_numbers (which we call bestmember and worstmember). 

[bestfitness(popcount),bestmember]=max(fitness);
[worstfitness(popcount),worstmember]=min(fitness);

% Next, save these (if want to save worstindividual can too)

bestindividual(:,popcount)=trait(:,bestmember,popcount);
%worstindividual(:,popcount)=trait(:,worstmember,popcount);

% Compute the average fitness in case you want to plot it.

avefitness(popcount) = sumfitness / POP_SIZE;        


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Create the next generation
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% First, form the mating pool.  
% To do this we select as parents the
% chromosomes that are most fit.  
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

   for pop_member = 1:POP_SIZE,
	if ELITISM ==1 & pop_member==bestmember  % If elitism on, and have
											 % the elite member
		parent_chrom(:,pop_member)=trait(:,pop_member,popcount); 
								  % Makes sure that
		                          % the elite member gets into the next
								  % generation.
	else
      pointer=rand*sumfitness; % This makes the pointer for the roulette 
                               % wheel.  
      member_count=1;        % Initialization
      total=fitness(1);

      while total < pointer,                % This spins the wheel to the 
       									    % pointer and finds the 
       									    % chromosome there - which is
       									    % identified by member_count
         member_count=member_count+1;
         total=total+fitness(member_count);
      end
      
% Next, make the parent chromosome

    parent_chrom(:,pop_member)=trait(:,member_count,popcount);
	end
   end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Reproduce section (i.e., make off-spring - "children")
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% In the approach below each individual gets to mate and
% they randomly pick someone else (not themselves) in the 
% mating pool to mate with.  Resulting children are 
% composed of a combination of genetic material of their
% parents.  For the AGA crossover is different since
% we have not split the numbers up into sequences of 
% digits.  We can, however, achieve essentially the same effect as
% the generation of a random cross site by the random
% generation of how much of each trait will be swapped
% with the corresponding trait from the mate and by
% randomly selecting a trait that is crossed over.
% Next, notice that if elitism is on, when the elite 
% member gets a chance to mate they do not; they are 
% simply copied over to the next generation.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

for parent_number1 = 1:POP_SIZE,    % Crossover (parent_number1 is the 
									% individual who gets to mate
	if ELITISM ==1 & parent_number1==bestmember % If elitism on, and 
											 % have the elite member
		child(:,parent_number1)=parent_chrom(:,parent_number1);
	else
	   parent_number2=parent_number1;  % Initialize who the mate is
	   while parent_number2 == parent_number1   % Iterate until find 
		                                        % a mate other than
												% yourself
         parent_number2 = rand*POP_SIZE; % Choose parent number 2
		 								 % randomly (a random mate)
         parent_number2 = parent_number2-rem(parent_number2,1)+1;     
	   end
    if CROSS_PROB > rand           % If true then crossover occurs
		                           % and here we simply then crossover
								   % all the traits on an individual
								   
	   alpha=rand;         % Specifies a random proportion of genetic
	                       % material that is going to be kept
						   % by parent_number1 in mating

       trait_site = rand*NUM_TRAITS; % Pick the trait that 
		 							 % is crossed over (i.e., where the split is made
       trait_site = trait_site-rem(trait_site,1)+1;     

% The next lines form the child by the swapping of genetic
% material between the parents.  

child(trait_site,parent_number1)=...
            alpha*parent_chrom(trait_site,parent_number1)+...
            (1-alpha)*parent_chrom(trait_site,parent_number2);
				% This takes part of parent_number1 (as given by 
				% alpha) and part of parent_number2 (specified
				% by 1-alpha, the remaining portion)
				% and makes a child from this
child(1:trait_site-1,parent_number1)=...
            parent_chrom(1:trait_site-1,parent_number1);
child(trait_site:NUM_TRAITS,parent_number1)=...
            parent_chrom(trait_site:NUM_TRAITS,parent_number2);
				% These last two lines make the child have the first
				% part of parent1 and the last part of parent2
				% In this way, at the trait site we think of randomly
				% splitting the two traits, and the child gets the
				% first traits of parent1 and the last traits of parent2
				% (clearly there are many other ways to do this - here we
				% simply pick one way to make it analogous to the standard
				% GA)
				
% Note that it is easy to modify the above code so that
% all traits are crossed over in different proportions. This 
% would then be a "multiple crossover point" strategy.
% One way to do this would be to simply let
%child(:,parent_number1)=...
%            alpha*parent_chrom(:,parent_number1)+...
%            (1-alpha)*parent_chrom(:,parent_number2);

			
% This crossover operation can result in traits that are out
% of the range specified by HIGHTRAIT and LOWTRAIT so we 
% put them back in range here.

if child(trait_site,parent_number1)>HIGHTRAIT(trait_site)

                     % The trait has went higher than the upper
                     % bound so let the trait equal to the
                     % HIGHTRAIT bound.

child(trait_site,parent_number1)=HIGHTRAIT(trait_site);

% Now consider the other case:

elseif child(trait_site,parent_number1)<LOWTRAIT(trait_site)

                     % The trait has went lower than the lower 
                     % bound so let the trait equal to the
                     % LOWTRAIT bound

child(trait_site,parent_number1)=LOWTRAIT(trait_site);
                  
end        


      else      % No crossover occurs

% Copy non-crossovered chromosomes into next generation
% In this case we simply take one parent and make them
% the child.

child(:,parent_number1)=parent_chrom(:,parent_number1);

	 end % Ends if CROSS_PROB... statement
   end  % End the "if ELITISM..." statement
end  % End "for parent_number1=..." loop

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Mutate children.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Here, we mutate to a different trait
% with a probability MUTAT_PROB.  Notice
% that this achieves a different, but similar
% effect to the standard approach to mutation.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

  for pop_member= 1:POP_SIZE,

	if ELITISM ==1 & pop_member==bestmember  % If elitism on, and 
											 % have the elite member
		child(:,pop_member)=child(:,pop_member); % Do not mutate
											% the elite member
	else

	for current_trait = 1:NUM_TRAITS,

		if MUTAT_PROB > rand        % If true then mutate  
            
            % To mutate simply change the trait to any allowable
			% one on the search space (hence this is a "trait mutation"
			% rather than a "gene mutation"

 		child(current_trait,pop_member)=...
		(rand-(1/2))*(HIGHTRAIT(current_trait)-LOWTRAIT(current_trait))+... 
		(1/2)*(HIGHTRAIT(current_trait)+LOWTRAIT(current_trait));
        end  % End "if MUTAT_PROB > rand ...                      
    end   
	end  % End "if ELITISM..."
  end  % End for pop_member loop

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Create the next generation (this completes the main part of the GA)
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

   popcount=popcount+1;		% Increment to the next generation
  			
   trait(:,:,popcount)=child;  % Create next generation (children 
   							   % become parents)       

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Terminate the program when the best fitness has not changed
% more than EPSILON over the last DELTA generations.  It would also
% make sense to use avefitness rather than bestfitness in this test.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

   if popcount > DELTA+1 & ...
       max(abs(bestfitness(popcount-DELTA:popcount-1)-...
          bestfitness(popcount-DELTA-1:popcount-2)))<=EPSILON
       break;
   end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
end  % End "for pop_count=..." loop - the main loop.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Next, provide some plots of the results of the simulation.
% These are for the example function that we seek to maximize; however
% for other applications it is not difficult to modify this code
% to show plots that are informative for the application at hand.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

t=1:popcount-1;  % For use in plotting

x=0:31/100:30;   % For our function the range of values we are considering
y=x;

% Compute the function that we are trying to find the maximum of.

for jj=1:length(x)
	for ii=1:length(y)
		z(ii,jj)=...
		+5*exp(-0.1*((x(jj)-15)^2+(y(ii)-20)^2))...
		-2*exp(-0.08*((x(jj)-20)^2+(y(ii)-15)^2))...
		+3*exp(-0.08*((x(jj)-25)^2+(y(ii)-10)^2))...
		+2*exp(-0.1*((x(jj)-10)^2+(y(ii)-10)^2))...
		-2*exp(-0.5*((x(jj)-5)^2+(y(ii)-10)^2))...
		-4*exp(-0.1*((x(jj)-15)^2+(y(ii)-5)^2))...
		-2*exp(-0.5*((x(jj)-8)^2+(y(ii)-25)^2))...
		-2*exp(-0.5*((x(jj)-21)^2+(y(ii)-25)^2))...
		+2*exp(-0.5*((x(jj)-25)^2+(y(ii)-16)^2))...
		+2*exp(-0.5*((x(jj)-5)^2+(y(ii)-14)^2));
	end
end

% First, show the actual function to be maximized and its contour map

figure(1)
clf
surf(x,y,z);
colormap(jet)
% Use next line for generating plots to put in black and white documents.
colormap(gray);
xlabel('x');
ylabel('y');
zlabel('z');
title('Fitness function');

figure(2) 
clf
contour(x,y,z,25)
colormap(jet)
% Use next line for generating plots to put in black and white documents.
colormap(gray);
xlabel('x');
ylabel('y');
zlabel('z');
title('Fitness function (contour map)');

hold on

% Next, add on the evolution of the other population
% members over time.

for pop_member = 1:POP_SIZE
	xx=squeeze(trait(1,pop_member,:));
	yy=squeeze(trait(2,pop_member,:));
	traitplot=plot(xx,yy,'k.');
	set(traitplot,'MarkerSize',6);
end

% Then, add on the evolution of the best individual over 
% time that the GA finds.

btraitplot=plot(bestindividual(1,:),bestindividual(2,:),'ko');
set(btraitplot,'MarkerSize',4)

hold off

% Next, plot some data on the operation of the GA

figure (3)
clf
subplot(211)
plot(t,bestfitness,'k--',t,avefitness,'k-',t,worstfitness,'k-.')
xlabel('Generation')
ylabel('Best, average, and worst fitness')
title('Best and worst fitness vs. generation')
subplot(212)
plot(t,bestindividual(1,:),'k-',t,bestindividual(2,:),'k--')
xlabel('Generation')
ylabel('Best individuals')
title('Best individuals vs. generation')
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% End of program
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%