p=[1103 6446;1190 7098;1313 7649;1491 8128;1668 8972;1871 9641;2186 10394;2371 11359;2750 12780]; 	 
t=[1483;1842;1804;1850;1993;2096;2713;3256;4049];
for i=1:3
net=newrb(p,t,0.01,i,6,2);
y(i,:)=sim(net,p);
end
figure;
plot(1:9,y(1,:)-t);
hold on;
plot(1:9,y(2,:)-t,'+');
hold on;
plot(1:9,y(3,:)-t,'*');
hold off;


p=[5,10,15,20,25,30,35,40,45,50,55,60]; 
t=[0.05,0.14,0.254,0.360,0.45,0.54,0.62,0.69,0.75,0.81,0.87,0.92]; 
goal=0.01; %誤差 
sp=1; %擴(kuò)展常數(shù) 
mn=100;%神經(jīng)元的最多個(gè)數(shù) 
df=1; %訓(xùn)練過(guò)程的顯示頻率 
[net,tr]=newrb(p,t,goal,sp,mn,df); %創(chuàng)建一個(gè)徑向基函數(shù)網(wǎng)絡(luò) 

a=sim(net,p); 

plot(p,t,'.-','markersize',20,'color',[1 0 0])% 

訓(xùn)練的很好 但在預(yù)測(cè)時(shí)就不行了！ 

不知道我的程序有什么問(wèn)題！ 

qq：9777852 


%輸入向量
p=[97 28 12 32 18 3 0 3 10 3 6 14 8 20 2 11 24 0 9 6 3 5 3 1 3 14 2 10 2 173 17 5 1 61 7;
66 29 11 34 13 3 0 1 14 4 12 11 8 12 2 13 20 0 4 3 4 5 4 1 0 12 0 9 3 250 4 3 0 58 3;
61 30 12 35 6 2 0 1 11 7 5 9 11 17 2 3 22 2 4 3 5 5 4 2 0 16 5 12 4 237 11 4 1 59 2;
79 28 18 43 7 6 1 5 9 6 6 20 17 30 1 12 20 0 9 5 2 5 5 3 2 16 5 14 3 249 20 3 0 56 3;
94 38 15 39 9 4 1 3 13 2 12 15 10 27 2 17 16 2 4 3 4 9 4 2 0 20 7 10 0 230 37 2 0 68 6;
106 53 16 44 17 3 0 2 15 3 18 12 7 19 11 0 32 0 4 9 2 6 6 3 2 8 0 10 4 331 25 1 1 104 5;
106 55 18 21 17 3 1 2 14 3 11 16 11 29 1 13 12 0 3 8 3 10 7 3 5 5 7 7 1 529 50 7 1 141 0;
113 75 19 29 42 9 2 3 26 0 13 21 5 26 2 5 27 1 5 20 2 5 18 4 4 1 3 17 5 643 30 2 7 120 4;
110 62 18 30 18 6 0 5 17 1 18 10 14 24 1 6 18 0 3 14 6 6 24 5 5 11 6 18 4 573 42 2 4 122 4;
114 93 23 28 29 6 0 9 23 4 8 25 15 36 0 4 28 3 5 16 4 2 36 3 2 7 9 12 7 699 54 8 8 139 8;
152 107 23 48 50 15 2 4 29 4 20 29 13 40 1 7 42 3 7 28 4 18 50 6 3 13 3 23 3 785 30 7 5 207 10]'; 

%目標(biāo)向量
t=[79 28 18 43 7 6 1 5 9 6 6 20 17 30 1 12 20 0 9 5 2 5 5 3 2 16 5 14 3 249 20 3 0 56 3;
94 38 15 39 9 4 1 3 13 2 12 15 10 27 2 17 16 2 4 3 4 9 4 2 0 20 7 10 0 230 37 2 0 68 6;
106 53 16 44 17 3 0 2 15 3 18 12 7 19 11 0 32 0 4 9 2 6 6 3 2 8 0 10 4 331 25 1 1 104 5;
106 55 18 21 17 3 1 2 14 3 11 16 11 29 1 13 12 0 3 8 3 10 7 3 5 5 7 7 1 529 50 7 1 141 0;
113 75 19 29 42 9 2 3 26 0 13 21 5 26 2 5 27 1 5 20 2 5 18 4 4 1 3 17 5 643 30 2 7 120 4;
110 62 18 30 18 6 0 5 17 1 18 10 14 24 1 6 18 0 3 14 6 6 24 5 5 11 6 18 4 573 42 2 4 122 4;
114 93 23 28 29 6 0 9 23 4 8 25 15 36 0 4 28 3 5 16 4 2 36 3 2 7 9 12 7 699 54 8 8 139 8;
152 107 23 48 50 15 2 4 29 4 20 29 13 40 1 7 42 3 7 28 4 18 50 6 3 13 3 23 3 785 30 7 5 207 10;
111 124 15 31 72 14 0 2 27 2 12 20 7 29 0 42 17 2 7 11 2 8 12 5 6 4 3 23 1 703 25 4 5 170 4;
144 116 24 52 52 12 1 4 12 3 11 28 14 45 2 19 18 0 5 17 6 11 14 4 13 15 5 31 3 674 24 6 9 166 7]' ; 

%歸一化
u=t;
%%%%%%歸一到 0 1 之間
for i=1:35
  p(i,:)=(p(i,:)-min(p(i,:)))/(max(p(i,:))-min(p(i,:)));
end

for i=1:35
  t(i,:)=(t(i,:)-min(t(i,:)))/(max(t(i,:))-min(t(i,:)));
end
%訓(xùn)練
%%%%%%%%%%threshold的作用是限定輸入的范圍，因?yàn)橐呀?jīng)做過(guò)歸一化，所以輸入的范圍在0 1之間
%%%%%%%%%%這個(gè)范圍比輸入的范圍大是可以的，舉個(gè)例子，比方說(shuō)歸一化的數(shù)據(jù)是在[0.1 0.9]之間
%%%%%%%%%%范圍用[0 1]是可以的，比歸一化后的范圍如果小的話(huà)，就會(huì)出問(wèn)題
threshold=[0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1; 0 1];
net=newff(threshold,[19,35],{'tansig','logsig'},'trainlm');
%%%%%%%%%%19隱層數(shù) 隱層數(shù)的確定有幾個(gè)經(jīng)驗(yàn)公式，也可以用試算法
net.trainParam.epochs=1000;
%訓(xùn)練次數(shù)
net.trainParam.goal=0.001;
%訓(xùn)練結(jié)束的目標(biāo)
LP.lr=0.1;
%這個(gè)應(yīng)該是學(xué)習(xí)率
net.trainParam.show=20;
P=[p(:,1)+p(:,2)+p(:,3) p(:,2)+p(:,3)+p(:,4) p(:,3)+p(:,4)+p(:,5) p(:,4)+p(:,5)+p(:,6) p(:,5)+p(:,6)+p(:,7) p(:,6)+p(:,7)+p(:,8) p(:,7)+p(:,8)+p(:,9) p(:,8)+p(:,9)+p(:,10) p(:,9)+p(:,10)+p(:,11)];
T=[t(:,1) t(:,2) t(:,3) t(:,4) t(:,5) t(:,6) t(:,7) t(:,8) t(:,9)];
%輸入樣本的輸入和目標(biāo)
net=train(net,P,T);
%訓(xùn)練網(wǎng)絡(luò)
test=[ t(:,7)+t(:,8)+t(:,9)];
out=sim(net,test); %仿真預(yù)測(cè)

%反歸一化
for i=1:35
  predict(i)=out(i)* (max(u(i,:))-min(u(i,:)))+ min(u(i,:));
end
predict

%結(jié)果整理
predictsum=predict(1);
sum=u(1,10);
for i=1:34
  predictsum= predictsum +predict(i+1);
  sum=sum+u(i+1,10);
end
  sum
predictsum
  sumerror= predictsum-sum

%繪制預(yù)報(bào)曲線(xiàn)
x=[t(:,10)];
a=x-out;
for i=1:35
  error=a* (max(u(i,:))-min(u(i,:)))+ min(u(i,:));
end
X=1:35;
figure(1);
plot(X,error)
xlabel('地上管網(wǎng)急修作業(yè)類(lèi)型（X）');
ylabel('作業(yè)量預(yù)測(cè)誤差（Y）');
figure(2);
plot(X, predict,'r*',X,u(:,10),'bo')
xlabel('地上管網(wǎng)急修作業(yè)類(lèi)型（X）');
ylabel('作業(yè)量（Y）');
text (23,predict(30), 'predicted data');
text(25.5,u(30,10), 'real data');