某地的CPI指标，d1、d2、d3、f为连续四年的指数，现在选取d1、d2、d3作为网络初始值，f为网络目标值进行训练网络。
具体数据如下：
d1=[0.2356 0.2589 0.3000 0.2445 0.2589 0.2366 0.2897 0.2368 0.2569 0.2010 0.2111 0.3012];
d2=[0.2562 0.3023 0.2454 0.2465 0.2146 0.2037 0.2898 0.2969 0.2110 0.2111 0.2212 0.3213];
d3=[0.3000 0.2458 0.2315 0.2216 0.2317 0.2218 0.2039 0.2510 0.2111 0.2212 0.2313 0.3114 ];
f=[0.2515 0.2146 0.2147 0.2218 0.2169 0.2260 0.2111 0.2212 0.2313 0.2414 0.2115 0.3216 ];
代码如下：
%%
clc;
clear;
% 数据输入
% 某地的CPI指标，有数据如下：
d1=[0.2356 0.2589 0.3000 0.2445 0.2589 0.2366 0.2897 0.2368 0.2569 0.2010 0.2111 0.3012];
d2=[0.2562 0.3023 0.2454 0.2465 0.2146 0.2037 0.2898 0.2969 0.2110 0.2111 0.2212 0.3213];
d3=[0.3000 0.2458 0.2315 0.2216 0.2317 0.2218 0.2039 0.2510 0.2111 0.2212 0.2313 0.3114 ];
f=[0.2515 0.2146 0.2147 0.2218 0.2169 0.2260 0.2111 0.2212 0.2313 0.2414 0.2115 0.3216 ];
k=1:1:12;
% 使用12个月的数据训练得到网络仿真计算相对误差
p1=zeros(1,12);
p2=zeros(1,12);
p3=zeros(1,12);
t=zeros(1,12);
p1(1)=d1(1);
for i=2:12
p1(i)=0.1*d1(i)+p1(i-1);
end
p2(1)=d2(1);
for i=2:12
p2(i)=0.1*d2(i)+p2(i-1);
end
p3(1)=d3(1);
for i=2:12
p3(i)=0.1*d3(i)+p3(i-1);
end
t(1)=f(1);
for i=2:12
t(i)=0.1*f(i)+t(i-1);
end
%数据标准化处理,标准化为网络输入p,期望输出tt
p=[(p1-mean(p1))./std(p1);(p2-mean(p2))./std(p2);(p3-mean(p3))./std(p3)];
tt=(t- mean(t))./std(t);
%用newff建立bp网络并训练
net=newff(minmax(p),[4,1],{'tansig','purelin'},'trainlm');
net.iw{1,1}=zeros(size(net.iw{1,1}))+0.5;
net.lw{2,1}=zeros(size(net.lw{2,1}))+0.75;
net.b{1,1}=zeros(size(net.b{1,1}))+0.5;
net.b{2,1}=zeros(size(net.b{2,1}));
net.trainParam.epochs=3000;
net.trainParam.goal =0.000005;
% 训练网络
net=train(net,p,tt);
%网络仿真得到网络输出，并计算误差
tt1=sim(net,p);
%利用标准化的逆变换得到t1的近似值to
% to为总增加率
to=tt1.*std(t)+mean(t);
a=zeros(1,12);
a(1)=to(1);
for i=2:12
a(i)=(to(i)-to(i-1))*10;
end
%累减得到近似的实际产值a
err_net=(tt1-tt)./tt;%网络相对误差
err_add=(to-t)./t;%总值累加数相对误差
err_real=(a-f)./f;%实际总值相对误差
%输出结果%
r1=[err_net;err_add;err_real]'%网络相对误差/累加数相对误差/实际值相对误差
r=err_real./err_add %误差放大倍数
l=[to;t;to-t;a;f;a-f]'%累加数计算值/累加数/绝对误差//计算的实际值/实际值/绝对误差
plotyy(k,to,k,t);%仿真累加值(to)随时间变化曲线与实际累加值(t)随时间变化曲线
% Create graphs with y-axes on both left and right side
% plotyy(X1,Y1,X2,Y2) plots X1 versus Y1 with y-axis labeling on the left
% and plots X2 versus Y2 with y-axis labeling on the right.
%
% This example graphs two mathematical functions using plot as the plotting function.
% The two y-axes enable you to display both sets of data on one graph
% even though relative values of the data are quite different. x = 0:0.01:20;
% y1 = 200*exp(-0.05*x).*sin(x);
% y2 = 0.8*exp(-0.5*x).*sin(10*x);
% [AX,H1,H2] = plotyy(x,y1,x,y2,'plot');
%
% You can use the handles returned by plotyy to label the axes
% and set the line styles used for plotting.
% With the axes handles you can specify the YLabel properties of the left-
% and right-side y-axis: set(get(AX(1),'Ylabel'),'String','Left Y-axis')
% set(get(AX(2),'Ylabel'),'String','Right Y-axis')
%
% Use the xlabel and title commands to label the x-axis and add a title:
% xlabel('Zero to 20 /musec.')
% title('Labeling plotyy')
%
% Use the line handles to set the LineStyle properties of the left-
% and right-side plots: set(H1,'LineStyle','--')
% set(H2,'LineStyle',':')
figure(2);
plotyy(k,a,k,f);%仿真值(a)随时间变化曲线与实际值(f)随时间变化曲线
