fnntrain.m

nnToolKit 神经网络工具包是基于 MATLAB 神经网络工具箱自行开发的一组神经网络算法函数库

function  retstr = FnnTrain(dt,ld,tt,sp)
  retstr=-1;
%%%% 输入参数赋值开始 %%%%%%%%%%%%%%%%%%%%%%%
% 方便调试程序用，程序调试时去掉这部分的注释
%
dt=4;       %学习阈值
ld=0.05;    %学习进度
tt=10;      %训练次数
sp='data\sample.txt';    %一个点的监测数据
%%%% 输入参数赋值结束 %%%%%%%%%%%%%%%%%%%%%%%%
  global recordDimention;
  global sampleNumber;
  global weightNumber;
  global distanceThread;
  global WW;
  global learningDegree;
  global epochsNumber;

  distanceThread=dt;
  learningDegree=ld;
  traintimes=tt;
  A=load(sp);
   
  [Arow Acol]=size(A);
  %样本个数
  sampleNumber=Arow;
  recordDimention=Acol;
  disp(sampleNumber);
  WW=A(1,:);
  WW=[WW [1]];
  weightNumber=1;
  epochsNumber=1;
  
  for jj=2:1:sampleNumber
      TrainNN2(A(jj,:));
  end
  for jt=1:traintimes-1
      for jt2=1:sampleNumber
          TrainNN2(A(jj,:));
      end
  end
 % 将训练结果写入权值文件
  dlmwrite('data\w.dat',WW,'\t');
%
% 训练子函数
%
function TrainNN2(a)
  global recordDimention;
  global sampleNumber;
  global weightNumber;
  global distanceThread;
  global WW;
  global learningDegree;
  global epochsNumber;
 
  Ldistance=zeros(2,weightNumber);
  for j1=1:weightNumber
      Ldistance(2,j1)=j1;
  end
  % %%%%%输入输出空间的模糊分割
  for j2=1:weightNumber
      Lx=0;
      for j3=1:recordDimention
          Lx=Lx + (a(j3) - WW(j2,j3)) .* (a(j3) - WW(j2,j3));
      end
      Ldistance(1,j2)=sqrt(Lx);
  end
  % %%%%%计算模糊空间的距离  
  Lx1=Ldistance(1,1);
  Lx2=Ldistance(2,1);
  for j4=2:weightNumber
      if(Ldistance(1,j4)<Lx1)
          Lx1=Ldistance(1,j4);
          Lx2=Ldistance(2,j4);
      end
  end
 % %%%%%修正模糊规则
  updateW=0;
  nowWeight=Lx2;
  if(Lx1<=distanceThread)
      for j6=1:recordDimention
          WW(nowWeight,j6)=WW(nowWeight,j6) + learningDegree .* (a(j6) - WW(nowWeight,j6));
      end
      WW(nowWeight,recordDimention+1)=WW(nowWeight,recordDimention+1)+1;
      updateW=1;
  end
    
  if( updateW==0)
      weightNumber=weightNumber+1;
      a=[a [1]];
      WW=[WW;a];
  end