bpsan.m

通过BP神经网络学习一个3维的函数的源代码

clear
x1=linspace(0.0001,2*pi,11);
y1=linspace(0.0001,2*pi,11);                %初始化输入值X1 Y1
[x1,y1]=meshgrid(x1,y1);        
d=(((sin(x1).*sin(y1))./(x1.*y1))+1)/2;     %初始化输出d
w1=[1.5844	1.6441;0.74448 0.78194;0.81343 -0.66575;-0.5737	0.83697;
    0.28679	0.44245;-0.57636 -1.22;0.80707	0.43714;-1.0981	-0.52348;
    0.088314 0.91418;1.5414	1.4782];
w2=[0.0026165 0.062183 -0.18009	0.13269	-0.80106 0.91097 0.66656 1.076 0.26821 -0.09642;
    1.469 2.1851 2.1371	2.1563 2.322 2.4231	2.0501 1.6318 2.1855 2.2569;
    0.38392	0.49159	-0.10762 0.39891 -1.1335 0.96899 -0.16392 1.0655 0.51813 -0.13409;
    -0.34162 0.35435 1.5207	1.2232 0.52976 0.40887	0.56438	0.97029	0.65146	-0.53335;
    1.7731	0.78681	0.58263	0.39657	0.30993	0.29067	0.76186	0.11722	0.47792	1.3413;
    0.34715	0.318 0.66788 0.013305	-1.3114	0.57609	-0.014608 0.82558 0.097232 0.14029;
    1.4849 2.3212 2.5465 1.779 1.5278 1.9225 1.8775	2.1854	1.9635	1.7167;
    0.088449 0.60285 -0.17105 0.14084 -1.5647 1.3829 0.75604 1.258	0.019266 0.57361;
    -0.037588 0.10421 1.332 1.0529 0.5272 0.4511 0.20397 0.7692	0.72259	0.64258;
    0.14971	0.85889	1.1346	0.33761	0.84096	0.67127	0.49248	0.82315	0.60204	0.46487];

w3=[0.334	1.2414	1.3603	0.66309	0.82581	1.2749	1.3164	0.9459	1.2181	1.1524;
    1.8235	-3.5995	0.8506	1.2803	1.0817	1.9224	-3.2308	1.695	1.8017	1.0148;
    0.35978	1.2114	1.3228	0.7178	0.29287	1.0126	0.96484	0.48741	1.0844	0.63083;
    1.1684	-3.5789	1.6871	1.04	1.3133	1.6784	-3.3743	1.4967	1.2023	1.7181;
    1.0875	1.2495	1.4182	0.70555	0.74038	1.236	1.288	1.1203	0.49184	0.70146;
    0.54223	-1.5256	0.90577	0.42566	1.5798	0.62715	-1.1558	0.75687	0.36601	0.29489;
    0.40877	1.4886	1.3601	0.93316	1.2761	0.70218	1.1033	0.58111	1.3082	1.2868;
    1.1168	-3.6942	1.116	1.7486	1.3142	1.1118	-2.7341	1.7122	1.2302	0.92108;
    1.4936	-2.8798	0.88438	0.89626	1.6421	1.1088	-2.5408	1.6672	0.73539	0.90525;
    0.6324	-1.8871	0.54249	0.93045	1.1186	0.74159	-2.158	1.3964	0.70555	0.76779];
w4=[2.9153	-6.1886	1.3527	-6.2903	2.7152	-3.3938	3.1009	-5.8818	-5.1855	-4.0715];
e=1.5;
step=0.0013;              %初始化权值w1 w2 w3 w4 误差e 学习率step

while e>0.007             %误差精度
x11=reshape(x1,121,1);
y11=reshape(y1,121,1);   %矩阵重置为121行1列
x=[x11';y11'];           %合并输入
    
s1=w1*x-1;
x2=1./(1+exp(-1*s1));
s2=w2*x2-1;
x3=1./(1+exp(-1*s2));
s3=w3*x3-1;
x4=1./(1+exp(-1*s3));
s4=w4*x4-1;
x5=1./(1+exp(-1*s4));     %计算输出

d11=reshape(d,121,1);
e=0.5*(d11'-x5)*(d11'-x5)';   %计算误差
disp(e);                       %显示e

o4=(d11'-x5).*x5.*(1-x5)*1;
i4=x4*o4';
w4=w4+i4'.*step;
o3=x4.*(1-x4).*(w4'*o4)*1;
i3=x3*o3';
w3=w3+1*i3'*step;
o2=x3.*(1-x3).*(w3'*o3)*1;
i2=x2*o2';
w2=w2+1*i2'*step;
o1=x2.*(1-x2).*(w2'*o2)*1;
i1=x*o1';
w1=w1+1*i1'*step;           %调整权值
end

xp=linspace(0.0001,2*pi,50);
yp=linspace(0.0001,2*pi,50);
[xp,yp]=meshgrid(xp,yp);
xp1=reshape(xp,2500,1);
yp1=reshape(yp,2500,1);
x=[xp1';yp1'];
s1=w1*x-1;
x2=1./(1+exp(-1*s1));
s2=w2*x2-1;
x3=1./(1+exp(-1*s2));
s3=w3*x3-1;
x4=1./(1+exp(-1*s3));
s4=w4*x4-1;
x5=1./(1+exp(-1*s4));
x5=reshape(x5,50,50);
mesh(xp,yp,x5);             %绘图程序
end