灰色神经网络

close all
clear
clc
% 16084
IN=1:42;
sr=zeros(1,42);
    sr(1)=1080.94;
    sr(2)=1083.03;
    sr(3)=1162.8;
    sr(4)=1155.61;
    sr(5)=1092.82;
    sr(6)=1099.26;
    sr(7)=1161.06;
    sr(8)=1258.05;
    sr(9)=1299.03;
    sr(10)=1298.3;
    sr(11)=1440.22;
    sr(12)=1641.3;
    sr(13)=1672.21;
    sr(14)=1612.73;
    sr(15)=1658.64;
    sr(16)=1752.42;
    sr(17)=1837.99;
    sr(18)=2099.29;
    sr(19)=2675.47;
    sr(20)=2786.33;
    sr(21)=2881.07;
    sr(22)=3183.98;
    sr(23)=3841.27;
    sr(24)=4109.65;
    sr(25)=3820.7;
    sr(26)=4471.03;
    sr(27)=5218.83;
    sr(28)=5552.3;
    sr(29)=5954.77;
    sr(30)=4871.78;
    sr(31)=5261.56;
    sr(32)=4383.39;
    sr(33)=4348.54;
    sr(34)=3472.71;
    sr(35)=3693.11;
    sr(36)=3433.35;
    sr(37)=2736.1;
    sr(38)=2775.72;
    sr(39)=2397.37;
    sr(40)=2293.78;
    sr(41)=1728.79;
    sr(42)=1871.16;
    OUT=sr;
  [X,minx,maxx,T,mint,maxt]=premnmx(IN,OUT);
q=50;
q1=0;
q0=70;
while(q1<q)
    q=q0;
    [M,N]=size(X);
    [L,N]=size(T);
    net=newff(minmax(X),[q,L],{'tansig','purelin'},'trainlm');
    net.trainParam.lr=0.05;
    net.trainParam.epochs=10000;
    net.trainParam.goal=1e-6;
    [net,tr]=train(net,X,T);
    Y=sim(net,X);
    Y=postmnmx(Y,mint,maxt);



    %灰色关联分析，调整网络隐层节点
    p=0.5;
    e=0.5;
    %此两个系数的设定是根据一些论文，已经实验的尝试得出的
    an=repmat(net.b{1},1,N);
    op=tansig(net.iw{1,1}*X+an);
    op1=op';
    T0=T';
    T1=repmat(T0,1,q);
    DIF=abs(T1-op1);
    MIN=min(min(DIF));
    MAX=max(max(DIF));
    Si=(MIN+p*MAX)./(DIF+p*MAX);
    ri=sum(Si)/N;
    D=find(ri>=e);
    [q0,q1]=size(D);
    q0=q1;
end
q0;
ri;
D;
q=q1;

%进行测试

PRD=1:42;
PRD=PRD';
P=tramnmx(PRD,minx,maxx);
TNEW=sim(net,P');
TNEW=postmnmx(TNEW,mint,maxt);


t=1:42;
YY=OUT;
YC=TNEW;
figure
plot(t,YY,'r*',t,YC,'b--+')
RES0=YC-YY;
res0=RES0./YY;
figure
bar(t,res0)