MATLAB 神经网络算法

[m0_46539152]

clear all
P=[1 1 1 ;1 1 1 ;1 1 0 ;1 1 1 ;1 1 0 ;1 1 1 ;1 1 0 ;1 1 0 ;1 1 0 ;1 1 0 ;1 1 1 ;1 1 1 ;1 1 1 ;1 1 0 ;1 1 0 ;]';
T=[1 1;1 1;0 0;1 1;0 0;1 1;0 0;0 0;0 0;0 0;1 1;1 1;1 1;0 0;0 0;]';
Px=[31.9 31.6 0.3 ;47.3 46.8 0.5 ;48.3 49.3 -1.0 ;51.9 51.6 0.3 ;80.6 81.3 -0.7 ;81.0 80.5 0.5 ;83.8 83.8 0.0 ;102.9 103.3 -0.4 ;108.3 108.3 0.0 ;113.4 113.9 -0.5 ;120.0 119.7 0.3 ;132.3 131.4 0.9 ;145.4 145.2 0.2 ;149.0 149.0 0.0 ;159.6 159.6 0.0 ;]';
Tx=[1.0 1.0 ;1.0 1.0 ;0.0 0.0 ;1.0 1.0 ;0.0 0.0 ;1.0 1.0 ;0.0 0.0 ;0.0 0.0 ;0.0 0.0 ;0.0 0.0 ;1.0 1.0 ;1.0 1.0 ;1.0 1.0 ;0.0 0.0 ;0.0 0.0 ;0.0 0.0]';
%net=newcf(minmax(P),[5,2],{'tansig','purelin'},'traingd');%创建两层前向反馈网络
%net=cascadeforwardnet(5,'traingdm');
net=feedforwardnet(5,'traingdm');

net.trainParam.epochs=10000;     %初始化训练次数
net.trainParam.goal=1e-020;      %初始化误差值
net.trainParam.lr = 0.5; 
[net,tr]=train(net,P,T);         %训练网络
Yout=sim(net,Px);
Tx1=Tx(1,:);
Tx2=Tx(2,:);
Yout1=Yout(1,:);
Yout2=Yout(2,:);
plot(Tx1);
hold on
plot(Yout1);
figure
plot(Tx2);
hold on
plot(Yout2);










clear all
close all
xite=0.50;
alfa=0.50;   %设置初始值

w2=rands(5, 2);   %设置输出层权值变量赋值
w2_1=w2 ;w2_2=w2_1;

w1=rand(3,5);    %隐含层权值变量赋值
w1_1=w1;w1_2=w1;
dw1=0*w1;
I=[0,0,0,0,0]';
Iout=[0,0,0,0,0]';
FI=[0,0,0,0,0]';


OUT=2;
k=0;
E=1.0;
NS=3;


while E>=1e-020      %控制训练结束的条件 误差最小
    k=k+1;
    times(k)=k;
    
 for s=1:1:NS        %MIMO samples
     xs=[1,0,0;0,1,0;0,0,1];
     ys=[1 0; 0.5 0.5;0,1];
     
     
     
     x=xs(s,:);
 for j=1:1:5
     I(j)=x*w1(:,j);               %计算隐含层的输出
     Iout(j)=1/(1+exp(-I(j)));     %计算输出层输出
 end
 
 
 
 y1=w2'*Iout;
 y1=y1' ;
 e1=0;
 y=ys(s,:);
 for l=1:1:OUT
     e1=e1+0.5*(y(1)-y1(1))^2;    %输出误差
 end
 es(s)=e1;
 E=0;
 if s==NS
     for s = 1:1:NS
         E=E+es(s);               %批量学习的误差
     end
 end
 ey=y-y1;
 
 w2=w2_1+xite*Iout*ey+alfa*(w2_1-w2_2); %带有动能因子的权值调解
 for j = 1:1:1:5
     S=1/(1+exp(-I(j)));
     FI(j)=S*(1-S);
 end
 for i=1:1:3
     for j=1:1:5
         dw1(i,j)=xite*FI(j)*x(i)*(ey(1)*w2(j,1)+ey(2)*w2(j,2));      %负梯度下降法
     end
 end
 w1=w1_1+dw1+alfa*(w1_1-w1_2);
 
 w1_2=w1_1;w1_1=w1;
 w2_2=w2_1;w2_1=w2;
 end                            % end of for
 Ek(k)=E;
end
figure(1);
plot(times,Ek,'r');
xlabel('k');ylabel('E');

x=[1 1 1 ;1 1 1 ;1 1 0 ;1 1 1 ;1 1 0 ;1 1 1 ;1 1 0 ;1 1 0 ;1 1 0 ;1 1 0 ;1 1 1 ;1 1 1 ;1 1 1 ;1 1 0 ;1 1 0 ;];
for i=1:1:7
    for j=1:1:5
    I(i,j)=x(i,:)*w1(:,j);
    Iout(i,j)=1/(1+exp(-I(i,j)));
    end
end
y=w2'*Iout';
Tx=[1.0 1.0 ;1.0 1.0 ;0.0 0.0 ;1.0 1.0 ;0.0 0.0 ;1.0 1.0 ;0.0 0.0 ;0.0 0.0 ;0.0 0.0 ;0.0 0.0 ;1.0 1.0 ;1.0 1.0 ;1.0 1.0 ;0.0 0.0 ;0.0 0.0 ;0.0 0.0];
Tx1=Tx(1,:);
Tx2=Tx(2,:);
y1=y(1,:);
y2=y(2,:);
plot(Tx1);
hold on
plot(y1);
figure
plot(y1);
figure
plot(Tx2);
hold on
plot(y2);