BP神經網路的預測，根據輸出層的是線性函式，這樣可以擬合出輸出和輸入之間的函式關係，

從而實現輸入新的值時，可以預測值，

例項：

<span style="font-size:18px;"><span style="font-size:18px;">clear;
p0=[11.9850 12.1121 12.2389 12.3626 12.4810 12.5768 12.6743 12.7627 12.8453 12.9227 12.9905 13.0756];

for i =1:7
    p(:,i)=[p0(i) p0(i+1) p0(i+2)]';
    t(i)=p0(i+3);
end
p
t

net=newff(minmax(p),[5,1],{'logsig','purelin'},'trainlm'),
net.trainParam.show = 100,
net.trainParam.epoch = 2000,
net.trainParam.goal= 1e-4,
[net,tr]=train(net,p,t);

ptest(:,1)=[p0(8) p0(9) p0(10)]';
ptest(:,2)=[p0(9) p0(10) p0(11)]';
ttest(1)=p0(11);
ttest(2)=p0(12);

result_test = sim(net,p)
result_test1 = sim(net,ptest)
delta = result_test1-ttest

result = [result_test result_test1];
plot([1997:2005],p0(4:12),'-r*',[1997:2005],result,'-o');

</span></span>
 

第二個例項：對下一個值進行預測

<span style="font-size:18px;">clear;

p0 =[123.46, 155.54, 204.38, 290.94, 462.71, 688.19, 988.85, 1342.04, 1773.29, 2431.21, 3330.82, 4792.7, 4556.26, 5695.8, 7019.79, 9712.29, 12028.54];

a=max(p0);
b=min(p0);
for i=1:17
p0(i) = (p0(i)-b)/(a-b);
end
%進行歸一化


for i = 1:11
    p(:,i)=[p0(i); p0(i+1); p0(i+2); p0(i+3)];
    t(i) = p0(i+4);
end

%輸入輸出的賦值
net=newff(minmax(p),[8,1],{'logsig','purelin'},'trainlm'),%建立神經網路
net.trainParam.show = 100,%練顯示的間隔
net.trainParam.epoch = 2000,%最大訓練次數
net.trainParam.goal= 1e-3,%訓練的最小均方誤差
[net,tr]=train(net,p,t);

%對神經網路進行訓練

ptest(:,1)=[p0(12) p0(13) p0(14) p0(15)]';
ptest(:,2)=[p0(13) p0(14) p0(15) p0(16)]';
ttest(1)=p0(16);
ttest(2)=p0(17);
%形成預測的資料和理想輸出

result_test = sim(net,p)
result_test1 = sim(net,ptest)
delta = result_test1-ttest%理想與預期的誤差

result = [result_test result_test1];</span>
 

clear;
p0=[1258.02 1267.43 1294.54 1298.66 1334.43 1354.66 1392.62 1416.8 1542.83 ...
    1635.86 1611.7 1660.15 1619.23 1548.22 1595.7 1651.22 1713.81 1708.22...
    1666.55 1675.05 1597.08 1583.72 1597.37 1614.59 1650.4 1664.58 1694.19...
    1733.39 1731.1];
a=max(p0);
b=min(p0);
for i=1:29
    p0(i)=(p0(i)-b)/(a-b);
end
for i = 1:22
    p(:,i)=[p0(i); p0(i+1); p0(i+2); p0(i+3)];
    t(i) = p0(i+4);
end


net=newff(minmax(p),[12,1],{'logsig','purelin'},'trainlm'),
net.trainParam.show = 100,
net.trainParam.epoch = 2000,
net.trainParam.goal= 1e-4,
[net,tr]=train(net,p,t);



ptest(:,1)=[p0(23) p0(24) p0(25) p0(26)]';
ptest(:,2)=[p0(24) p0(25) p0(26) p0(27)]';
ptest(:,3)=[p0(25) p0(26) p0(27) p0(28)]';
ttest(1)=p0(27);
ttest(2)=p0(28);
ttest(3)=p0(29);
result_test = sim(net,p);
result_test1 = sim(net,ptest)
delta = result_test1-ttest
result = [result_test result_test1];
plot([1:25],p0(5:29),'-r*',[1:25],result,'-o');