遗传rbf网络.m

基于遗传算法对rbf神经网络的径向基宽度进行优化的matlab程序

%种群个体数目 20  nind  变量数 1  nvar
echo off
clear
clc
NIND = 20         %个体数目
NVAR = 1;         %变量维数
Range = [0;5];    %变量范围
GGAP = .8;           % Generation gap, how many new individuals are created。代沟
INSR = .9;           % Insertion rate, how many of the offspring are inserted。插入率
XOVR =  1;           % Crossover rate                                          交叉率
SP = 2;              % Selective Pressure                                      
SUBPOP = 12          % Number of subpopulation                                 子种群数
MUTR = 1;            % Mutation rate; only a factor;                            变异概率
MIGR = 0.2;          % Migration rate between subpopulations                    迁移率
MIGGEN = 20;         % Number of generations between migration (isolation time)  在子种群与迁移之间20代

MAXGEN = 500 ; % Maximal number of generations                                 最大遗传代数

TERMEXACT = 1e-4;    % Value for termination if minimum reached

SEL_F = 'sus';       % Name of selection function
XOV_F = 'recdis';    % Name of recombination function for individuals
MUT_F = 'mutbga';    % Name of mutation function

FieldDD=rep(Range,[1,NVAR])   % 每个变量的范围

OBJ_F = 'rbfrmse';   % Name of function for objective values

 
% Create real population
   Chrom = crtrp(SUBPOP*NIND,FieldDD);

   trace=zeros(MAXGEN,2);
% reset count variables
   gen = 0;
  
   % Calculate objective function for population
   ObjV = feval(OBJ_F,Chrom);
   % count number of objective function evaluations
  
% Iterate subpopulation till termination or MAXGEN
   while ((gen < MAXGEN) ),

    % Fitness assignement to whole population
      FitnV = ranking(ObjV,[2 0],SUBPOP);
            
   % Select individuals from population
      SelCh = select(SEL_F, Chrom, FitnV, GGAP, SUBPOP);
      
   % Recombine selected individuals
      SelCh=recombin(XOV_F, SelCh, XOVR, SUBPOP);

   % Mutate offspring
      SelCh=mutate(MUT_F, SelCh, FieldDD, [MUTR], SUBPOP);

   % Calculate objective function for offsprings
      ObjVOff = feval(OBJ_F,SelCh);
      
   % Insert best offspring in population replacing worst parents
      [Chrom, ObjV] = reins(Chrom, SelCh, SUBPOP, [1 INSR], ObjV, ObjVOff);

      gen=gen+1;

      [trace(gen,1),I]=min(ObjV);
      trace(gen,2)=mean(ObjV);
      
   % Plot some results, rename title of figure for graphic output
      if ((rem(gen,MAXGEN) == 0) ),
         [Chrom, ObjV] = migrate(Chrom, SUBPOP, [MIGR, 1, 1], ObjV);
      end  
  end
[Y,I]=min(ObjV);
figure(1);plot(Chrom(I,:));
hold on; grid;
plot(Chrom(I,:),'bo');
% figure(2);plot(-trace(:,1));
% hold on;
% plot(-trace(:,2),'-.');
% legend(

 

function ObjVal=rbfrmse(Chrom)

load DATA01
load REALMFR1

[NIND,NVAR] = size(Chrom);
ObjVal=zeros(NIND,1);

Set=[a210a' f130a' f141a' t210' p210' l210' f211' f213' f217' realmfr1];
Gset=Set(77:85,:);
Set(77:85,:)=[];
VerifySet=Set(1:4:76,:);%验证集  19个样本
Set(1:4:76,:)=[];
TrainSet=Set;%训练集  57个样本

TrainX=TrainSet(:,1:9);
TrainY=TrainSet(:,10);

VerX=VerifySet(:,1:9);
VerY=VerifySet(:,10);

%训练
[pn meanp stdp]=prestd(TrainX');      %数据标准化
[pca,transmat]=prepca(pn,0.05);       %主元分析，提取95%的方差

%数据标准化
[R C]=size(VerX);
for i=1:R                          
pn2(:,i) = (VerX(i,:)'-meanp)./stdp;   
end

pca2=transmat*pn2;  %验证数据的主元提取     

for i=1:NIND
net=newrb(pca,TrainY',0.01,Chrom(i));      %训练神经网络

V=sim(net,pca2);   %验证数据的输出

ObjVal(i)=1/rmse(V-VerY');
end
function perf=rmse(e)
%
% calculate the root mean squared error of the given errors
% 
%  'perf = rmse(E);'
%
% see also:
%    mae, linf, trimmedmse
%

% Copyright (c) 2002,  KULeuven-ESAT-SCD, License & help @ http://www.esat.kuleuven.ac.be/sista/lssvmlab


perf = sqrt(sum(sum(e.^2)) / prod(size(e)));