BP神经网络程序问题求助

T-Eric

这是我从书上打下来的代码，因为书的运行软件版本比较低，我的是Malab2013a的版本，所以我修改了部分程序，但运行出来有问题。请哪位高手帮忙指教和修正一下。谢谢！！

%原始数据输入
5 G+ g4 U6 e# u7 R8 \8 f; Iclc
sqrs=[20.55,22.44,25.73,27.13,29.45,30.1,30.96,34.06,36.42,38.09,39.13,39.99,...
    41.93,44.59,47.30,52.89,55.73,56.76,59.17,60.63];
sqjdcs=[0.6,0.75,0.85,0.90,1.05,1.35,1.45,1.60,1.70,1.85,2.15,2.20,2.25,2.35,...
" ]4 P1 h9 g+ X: U8 @7 Y% s8 ?7 S    2.5,2.6,2.7,2.85,2.95,3.10];
: }7 _3 I5 |; B% ^( y! K6 Isqglmj=[0.09,0.11,0.11,0.14,0.20,0.23,0.23,0.32,0.32,0.34,0.36,0.36,0.38,0.49,...
- m0 c8 j$ {, C, Q    0.56,0.59,0.59,0.67,0.69,0.79];
glkyl=[5126,6217,7730,9145,10460,11387,12353,15750,18304,19836,21024,19490,20433,...
    22598,25107,33442,36836,40548,42927,43462];
. i- ?7 l: X% \$ e: ?2 hglhyl=[1237,1379,1385,1399,1663,1714,1834,4322,8132,8936,11099,11203,10524,11115,...
    13320,16762,18673,20724,20803,21804];
p=[sqrs;sqjdcs;sqglmj];
t=[glkyl;glhyl];
0 s2 j; Z2 K& v- Q% x1 \8 d" |
7 f+ E/ F. W( t8 ^/ p%数据归一化
[pn,ps1]=mapminmax(p);
[tn,ps2]=mapminmax(t);
. H, w4 y" E* m& r2 Ddx=[-1,1;-1,1;-1,1];

; b8 F( }! F' W7 S%BP网络训练
net=newff(p,t,[3,7,2],{'tansig','tansig','purelin'},'traingdx');
' ]) @# _( c  u4 L9 ]net.trainParam.show=1000;
net.trainParam.Lr=0.05;
* G' n/ Z; u6 e' Knet.trainParam.epochs=50000;
net.trainParam.goal=0.65*10^(-3);
net=train(net,pn,tn);
9 F+ W! I! l: Y( k
%利用原数据对BP网络仿真
( r- b- S- X1 gan=sim(net,pn);
a=mapminmax('reverse',an,ps2);

%仿真结果与原数据对比测试
1 S: k' [& h- _0 ]4 w- fx=1990:2009;
newk=a(1,:);
newh=a(2,:);
0 z. p, E, H! n# P9 p/ ~! Hfigure(2);
' G0 ]/ ~- [& b( qsubplot(2,1,1);plot(x,newk,'r-o',x,glkyl,'b--+');
& x! R3 z% L% [6 J, u. d# rlegend('网络输出客运量','实际客运量');
6 a) Q. \* k5 `$ T1 g7 Y' X% N& w; nxlabel('年份');ylabel('货运量、万人');
2 M+ Q* C# b' q) P* V/ Ntitle('运用工具箱客运量学习和测试对比图');
subplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');
* d  L/ P; O3 ~, e% v5 j1 }legend('网络输出货运量','实际货运量');
$ T6 K* a! G, P. L* E# g' zxlabel('年份');ylabel('货运量、万吨');
' T! ?+ a  ?4 a4 ititle('运用工具箱货运量学习和测试对比图');
9 X) W2 n$ S4 r
%新数据仿真
pnew=[73.39,75.55
4 n- l# r8 Q: X7 g& d8 g    3.9635,4.0975
    0.9880,1.0268];
) E5 l0 r( W% D3 fpnewn=mapminmax('apply',pnew,ps1);
6 B$ U% L5 N  o) A* |  G1 vanewn=sim(net,pnewn);
anew=mapminmax('reverse',anewn,ps2)
* J0 y* _0 c4 z
8 O6 T: F4 Y1 c  ^/ S8 s' c. {7 S

gancm

看了图我感觉是网络输出后面那几个值太大了，前面基本都是一条水平线了，具体哪里还看不出问题。你把修改的地方说一下看看

529084167

T-Eric

gancm 发表于 2014-7-19 12:30
* j5 q' L, q# |. z2 ~看了图我感觉是网络输出后面那几个值太大了，前面基本都是一条水平线了，具体哪里还看不出问题。你把修改的 ...

因为版本更新，Matlab中的Newff命令用法有所改变，原命令为：
net=newff(dx,[3,7,2]{'tansig','tansig','purelin'},'traingdx');
我修改为：
net=newff(p,t,[3,7,2],{'tansig','tansig','purelin'},'traingdx');
8 a& h, I3 s% O然后有些命令被其他命令替代了，其中有premnmx，postmnmx，tramnmx命令好像被替换了，使用了一个强大的命令mapminmax。原命令为：
[pn,minp,maxp,tn,mint,maxt]=premnmx(p,t);
5 W# T2 o& `! ma=postmnmx(an,mint,maxt);
pnewn=tramnmx(pnew,minp,maxp);
anew=postmnmx(anewn,mint,maxt)
我修改为：
[pn,ps1]=mapminmax(p);[tn,ps2]=mapminmax(t);
2 k6 b7 S. F" N' [a=mapminmax('reverse',an,ps2);
5 ~4 E: I$ F: Jpnewn=mapminmax('apply',pnew,ps1);
) P+ ?: z2 F# P& V& ]& L7 g2 eanew=mapminmax('reverse',anewn,ps2)

原程序为：
  X. ?! z9 O- A%原始数据输入
clc
sqrs=[20.55,22.44,25.73,27.13,29.45,30.1,30.96,34.06,36.42,38.09,39.13,39.99,...
    41.93,44.59,47.30,52.89,55.73,56.76,59.17,60.63];
sqjdcs=[0.6,0.75,0.85,0.90,1.05,1.35,1.45,1.60,1.70,1.85,2.15,2.20,2.25,2.35,...
7 [/ W- I+ ]: Z! c* j3 ]/ g    2.5,2.6,2.7,2.85,2.95,3.10];
sqglmj=[0.09,0.11,0.11,0.14,0.20,0.23,0.23,0.32,0.32,0.34,0.36,0.36,0.38,0.49,...
    0.56,0.59,0.59,0.67,0.69,0.79];
glkyl=[5126,6217,7730,9145,10460,11387,12353,15750,18304,19836,21024,19490,20433,...
6 G0 z0 ~4 d7 e; q' W; A- O    22598,25107,33442,36836,40548,42927,43462];
glhyl=[1237,1379,1385,1399,1663,1714,1834,4322,8132,8936,11099,11203,10524,11115,...
    13320,16762,18673,20724,20803,21804];
p=[sqrs;sqjdcs;sqglmj];
t=[glkyl;glhyl];

9 d( H8 s  B4 J+ ~# H%数据归一化
[pn,minp,maxp,tn,mint,maxt]=premnmx(p,t);
" k$ i6 g" r% E0 C/ Ydx=[-1,1;-1,1;-1,1];
: }6 Z/ \( B# C/ q
%BP网络训练
1 f- c" h4 x2 z+ n1 anet=newff(dx,[3,7,2],{'tansig','tansig','purelin'},'traingdx');
net.trainParam.show=1000;
net.trainParam.Lr=0.05;
net.trainParam.epochs=50000;
2 }; B) f' O) S* k. b' N" N5 hnet.trainParam.goal=0.65*10^(-3);
2 G5 J2 K. s9 p1 a/ J; snet=train(net,pn,tn);
9 U9 a& E) w' w0 G: N" A  C9 |0 w' A
5 M# A: s% ?. f%利用原数据对BP网络仿真
an=sim(net,pn);
a=postmnmx(an,mint,maxt);
' N! I7 D/ ?! o( L' j
%仿真结果与原数据对比测试
  a) ~7 D: U* P* Z  J- H3 E- D8 bx=1990:2009;
  z. n8 p& o; [, {newk=a(1,;
newh=a(2,;
* p" x& t  J1 l$ i6 t6 C4 sfigure(2);
: H6 Z; |6 z# c, msubplot(2,1,1);plot(x,newk,'r-o',x,glkyl,'b--+');
legend('网络输出客运量','实际客运量');
xlabel('年份');ylabel('货运量、万人');
title('运用工具箱客运量学习和测试对比图');
" d: g/ P" _  ~7 J) J3 x1 f; @subplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');
legend('网络输出货运量','实际货运量');
xlabel('年份');ylabel('货运量、万吨');
! o9 \- `) _+ H$ i; J, [title('运用工具箱货运量学习和测试对比图');
# S1 b! D' L2 D# o! O
6 s. g  q0 y' [%新数据仿真
& {& @) w" d9 k3 b" |pnew=[73.39,75.55
9 E" {- G* l$ k8 |2 d* a    3.9635,4.0975
6 l; `% ^$ n; x& k; t& h6 F    0.9880,1.0268];
! v- b+ O/ e# I2 W  Wpnewn=tramnmx(pnew,minp,maxp);
6 ?- N/ Q9 r: D6 l: m/ H( Z4 B$ \anewn=sim(net,pnewn);
5 M0 U; F+ C& C0 Q1 i6 m( g' t# yanew=postmnmx(anewn,mint,maxt)

3 o% V; \! A5 K/ ~. \* b2 }2 m修改后程序为：
, P" h- i! O! K2 ]* E' e%原始数据输入
& J% l  e7 n( Sclc
sqrs=[20.55,22.44,25.73,27.13,29.45,30.1,30.96,34.06,36.42,38.09,39.13,39.99,...
    41.93,44.59,47.30,52.89,55.73,56.76,59.17,60.63];
! e/ M* Q$ s' f+ l: n/ m9 ]  hsqjdcs=[0.6,0.75,0.85,0.90,1.05,1.35,1.45,1.60,1.70,1.85,2.15,2.20,2.25,2.35,...
    2.5,2.6,2.7,2.85,2.95,3.10];
' ?( [/ C% G& ^7 Z% b! N' o, \sqglmj=[0.09,0.11,0.11,0.14,0.20,0.23,0.23,0.32,0.32,0.34,0.36,0.36,0.38,0.49,...
    0.56,0.59,0.59,0.67,0.69,0.79];
( h4 C! F- \+ v+ ~. b* y6 b) V8 sglkyl=[5126,6217,7730,9145,10460,11387,12353,15750,18304,19836,21024,19490,20433,...
. w  _# x+ w' L% d8 Z, J1 j    22598,25107,33442,36836,40548,42927,43462];
glhyl=[1237,1379,1385,1399,1663,1714,1834,4322,8132,8936,11099,11203,10524,11115,...
    13320,16762,18673,20724,20803,21804];
# A; I, D$ `5 n# K+ @" Jp=[sqrs;sqjdcs;sqglmj];
t=[glkyl;glhyl];
/ _3 H" b) w% l7 g" }8 K
- ]  l4 @5 W0 G; V" @% U" x8 @%数据归一化
- j7 a* z6 a- y+ b+ D[pn,ps1]=mapminmax(p);
[tn,ps2]=mapminmax(t);

: R5 O! [1 P5 G# x%BP网络训练
net=newff(p,t,[3,7,2],{'tansig','tansig','purelin'},'traingdx');
net.trainParam.show=1000;
% }/ G: z2 w; g3 ~net.trainParam.Lr=0.05;
6 M) x3 x% [1 I; A0 Y: vnet.trainParam.epochs=50000;
, v. [" z9 P' s* `& }7 Rnet.trainParam.goal=0.65*10^(-3);
) E. b/ @2 T9 _, j9 [5 Onet=train(net,pn,tn);

: l/ {1 _" I7 ~%利用原数据对BP网络仿真
1 K# t; p8 ?$ H! N0 `4 oan=sim(net,pn);
" b$ Y, T3 }$ x. d8 x- n6 Ia=mapminmax('reverse',an,ps2);

2 b& R  }( Q8 Z9 w2 u%仿真结果与原数据对比测试
x=1990:2009;
4 W* m2 A/ p; L9 p* Q, T5 Enewk=a(1,;
newh=a(2,;
figure(2);
2 Z: R* P9 d, z+ S( n/ ^5 R$ osubplot(2,1,1);plot(x,newk,'r-o',x,glkyl,'b--+');
# x3 l6 J! ~% j! y2 e# N% ?legend('网络输出客运量','实际客运量');
xlabel('年份');ylabel('货运量、万人');
title('运用工具箱客运量学习和测试对比图');
& F+ l1 K+ Q$ ~8 g) lsubplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');
1 G* R) c3 O/ t: qlegend('网络输出货运量','实际货运量');
xlabel('年份');ylabel('货运量、万吨');
. c. Z! C- K' D2 E4 C& {9 xtitle('运用工具箱货运量学习和测试对比图');

%新数据仿真
, w, t" |3 O7 {, t; Y5 opnew=[73.39,75.55
    3.9635,4.0975
6 B: ]7 K9 Q/ n& G    0.9880,1.0268];
pnewn=mapminmax('apply',pnew,ps1);
1 F5 b0 ^+ H' r7 ^3 ]anewn=sim(net,pnewn);
0 {1 ^3 Q6 t. h" M; U. l- S+ ^# @anew=mapminmax('reverse',anewn,ps2)
) b  [# w# E1 I3 a( v（修改的地方用颜色标记了）
5 U$ n4 |6 V% Z# `8 G; F7 F麻烦您帮忙指出其中的问题，万分感谢！

且生

求书名及页码……

且生

1. clc
   3 F. Q, ~. v7 b, A6 e, s
2. sqrs=[20.55,22.44,25.73,27.13,29.45,30.1,30.96,34.06,36.42,38.09,39.13,39.99,.... D8 D6 T7 m1 R* i; X4 y7 e4 @\" D
   
3.     41.93,44.59,47.30,52.89,55.73,56.76,59.17,60.63];3 C; o6 R2 ~9 u6 f) b. e8 n% u
   
4. sqjdcs=[0.6,0.75,0.85,0.90,1.05,1.35,1.45,1.60,1.70,1.85,2.15,2.20,2.25,2.35,.../ i1 |; b3 L4 I
   
5.     2.5,2.6,2.7,2.85,2.95,3.10];. y/ x% Z4 s& Q' ?* z+ ^
   
6. sqglmj=[0.09,0.11,0.11,0.14,0.20,0.23,0.23,0.32,0.32,0.34,0.36,0.36,0.38,0.49,...% @# B+ c5 y; d
   
7.     0.56,0.59,0.59,0.67,0.69,0.79];; g% G1 ?5 X: W4 r7 D
   
8. glkyl=[5126,6217,7730,9145,10460,11387,12353,15750,18304,19836,21024,19490,20433,...
   4 ^+ O6 |) ?; _- P+ C7 V
9.      22598,25107,33442,36836,40548,42927,43462];
   - \0 X& h. d7 y( C
10. glhyl=[1237,1379,1385,1399,1663,1714,1834,4322,8132,8936,11099,11203,10524,11115,...* f* v! ^% a- a  Z; M
    
11.      13320,16762,18673,20724,20803,21804];% ?4 Z) z. N& z* P% w' N/ g2 D
    
12. p=[sqrs;sqjdcs;sqglmj];
    - n& d1 X5 z% M6 g
13. t=[glkyl;glhyl];
    8 N4 p6 x! K+ f. U, y' B; k
14. %数据归一化
    3 ]* G, M3 w) o% Q$ g
15. [pn,ps1]=mapminmax(p);' F0 K2 G. d  B1 Z; }9 w+ K' a
    
16. [tn,ps2]=mapminmax(t);
    . ?/ ]+ h$ S6 W  H0 Y
17. %dx=[-1,1;-1,1;-1,1];& D, N- H3 b/ s  t/ a
    
18. [color=Red] p0=minmax(pn);t0=minmax(tn);[/color]# K* b2 d* |, y
    
19. 
    9 u: E  G5 |$ l( [0 h
20. %BP网络训练
    0 D! B: T1 i4 d# ^6 j2 ^+ F/ ?
21. [color=Red] net=newff(p0,t0,[3,7,2],{'tansig','tansig','purelin'},'traingdx'); [/color]
    4 G1 O8 D' `4 |2 I2 A
22. net.trainParam.show=1000;; P( o: S$ H) s: p
    
23. net.trainParam.Lr=0.05;+ F& A- p1 M# a
    
24. net.trainParam.epochs=50000;& S0 v6 y/ t) ], |/ J2 P1 ~5 q
    
25. [color=Red] net.trainParam.goal=0.65*10^(-5);[/color]
    # [; _1 J/ l( D+ i+ b
26. net=train(net,pn,tn);1 i. @/ d7 L; _( d: g5 ~; N
    
27. 5 i- N$ U\" D' y+ u1 m& k
    
28. %利用原数据对BP网络仿真
    5 \6 w/ e1 L/ T
29. an=sim(net,pn);& O& W+ Z+ ?2 m3 k! M\" s
    
30. a=mapminmax('reverse',an,ps2);
    7 K& x+ ~$ ]( q. A) r) a9 z2 @( H
31. : O) g- n* v, l# \1 Y% C9 @, h  f6 f
    
32. %仿真结果与原数据对比测试 % p1 }5 L! b. {) R
    
33. x=1990:2009;
    7 M# A  F5 t* ]8 f5 u: k! k
34. newk=a(1,:);2 _! w! j' O0 A; _
    
35. newh=a(2,:);  |- m9 Y0 z; N4 G; V) _
    
36. figure(2);5 y& ]- [/ `+ z. u\" j; O8 ~: K
    
37. subplot(2,1,1);plot(x,newk,'r-o',x,glkyl,'b--+');6 s\" Q4 A- f7 X$ O# y% N- o
    
38. legend('网络输出客运量','实际客运量');0 J) k) n0 S0 c' \/ Z\" q- ~/ N  \
    
39. xlabel('年份');ylabel('货运量、万人');
      F& K2 w& ^\" o& C& z
40. title('运用工具箱客运量学习和测试对比图');% M1 d5 I% N5 C) l3 U; {3 L
    
41. subplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');* O  S2 t( O# o
    
42. legend('网络输出货运量','实际货运量');
    - I3 Q$ G( p% B' I/ j7 R
43. xlabel('年份');ylabel('货运量、万吨');: a& ~/ o% s3 T
    
44. title('运用工具箱货运量学习和测试对比图');- S5 z: C# I8 @! Q$ F3 r/ s. M3 k
    
45. 2 v6 E5 `. X7 ~2 M/ I9 j
    
46. %新数据仿真
    * e. Y9 N- o3 a6 h4 ]1 S4 f\" j
47. pnew=[73.39,75.55+ O5 x; `( G: F2 y  R4 N
    
48.      3.9635,4.09751 }) @# i$ E1 m7 A* {
    
49.      0.9880,1.0268];* P& q2 O: Z, e* [: d, G
    
50. pnewn=mapminmax('apply',pnew,ps1);
    , [# h  f8 t& p; h+ [+ {
51. anewn=sim(net,pnewn);
    0 n& D8 Q  M7 v4 b0 M* s
52. anew=mapminmax('reverse',anewn,ps2)

复制代码

且生

忽略我上面那个，改动的地方在18，21，25行……求批

1. clc\" i# X& w% G$ ^% D7 T$ m# _
   
2. sqrs=[20.55,22.44,25.73,27.13,29.45,30.1,30.96,34.06,36.42,38.09,39.13,39.99,...1 J7 z2 H2 w+ a) e$ q! p- {% r8 g
   
3.     41.93,44.59,47.30,52.89,55.73,56.76,59.17,60.63];+ N1 B+ [6 o$ u7 t$ R7 w
   
4. sqjdcs=[0.6,0.75,0.85,0.90,1.05,1.35,1.45,1.60,1.70,1.85,2.15,2.20,2.25,2.35,...
   / H) S3 `; r. B/ A/ n
5.     2.5,2.6,2.7,2.85,2.95,3.10];
   . e7 t; Y+ @$ Z! o& H\" U( H
6. sqglmj=[0.09,0.11,0.11,0.14,0.20,0.23,0.23,0.32,0.32,0.34,0.36,0.36,0.38,0.49,.... ~! ]: b2 |) Q  v, l( O) ~
   
7.     0.56,0.59,0.59,0.67,0.69,0.79];
   ) r6 l: P8 Q6 M, g\" Y4 v+ f9 Z
8. glkyl=[5126,6217,7730,9145,10460,11387,12353,15750,18304,19836,21024,19490,20433,...
   7 \3 C9 e3 Q7 K# w8 a9 ^
9.      22598,25107,33442,36836,40548,42927,43462];7 ^( `* U) N$ Z2 C, F2 R
   
10. glhyl=[1237,1379,1385,1399,1663,1714,1834,4322,8132,8936,11099,11203,10524,11115,...
    1 a7 ^8 r2 }. I
11.      13320,16762,18673,20724,20803,21804];
    - v1 M2 |0 q& [; T7 v; b! N
12. p=[sqrs;sqjdcs;sqglmj];* C0 W: n# Y. r2 d2 ^+ A
    
13. t=[glkyl;glhyl];
    2 b+ M6 U* ~8 `! l& y
14. %数据归一化: \2 [\" ?. Y2 ]8 A2 F9 D$ s& U& [
    
15. [pn,ps1]=mapminmax(p);\" [- t$ M0 E: B- w# X0 [
    
16. [tn,ps2]=mapminmax(t);
    # S; O1 z/ X\" n; m) P3 C
17. %dx=[-1,1;-1,1;-1,1];
    - E, e6 k# ?/ K. ?2 o
18. p0=minmax(pn);t0=minmax(tn);9 I2 d( Y! L, b- n4 b5 g. Q
    
19. ! L1 `7 {% a6 |. y9 c. @
    
20. %BP网络训练
    % ?1 w6 Z3 @* p, @( T$ J
21. net=newff(p0,t0,[3,7,2],{'tansig','tansig','purelin'},'traingdx');
    4 M! d- y- I6 Z3 h( p! B5 E/ ^
22. net.trainParam.show=1000;; s; P, d$ j0 H9 Q2 i
    
23. net.trainParam.Lr=0.05;: }$ j\" x% y' c% f( D
    
24. net.trainParam.epochs=50000;
      r  S6 ]* `! B- X
25. net.trainParam.goal=0.65*10^(-5);
    : n! z, k) q4 ^% E: F
26. net=train(net,pn,tn);! v! F5 O) U! R5 I, m& j' S\" ~
    
27. & q0 `( p; F4 [
    
28. %利用原数据对BP网络仿真: D! G2 q$ @( n\" N' W3 Y
    
29. an=sim(net,pn);5 u$ Z. i3 e/ e( M% c5 E
    
30. a=mapminmax('reverse',an,ps2);
    3 P  \$ M5 d0 Y( f8 j8 f
31. 
    7 L2 Z\" V& T! u! N5 g+ E2 S
32. %仿真结果与原数据对比测试 + n, X: [& D7 d2 v2 W6 E
    
33. x=1990:2009;
    1 \5 H4 n% A& F% o* I8 ?
34. newk=a(1,:);' P; Z; Q1 \5 @9 J- s
    
35. newh=a(2,:);
      S! H4 I8 I6 x2 d
36. figure(2);
    8 L+ U1 L' G& s6 q  c$ q, V
37. subplot(2,1,1);plot(x,newk,'r-o',x,glkyl,'b--+');' {# l* ]- Y7 y! [2 X
    
38. legend('网络输出客运量','实际客运量');* D+ E/ X8 _+ v. _/ T8 Y' s; m
    
39. xlabel('年份');ylabel('货运量、万人');' A; ~4 z+ J4 c
    
40. title('运用工具箱客运量学习和测试对比图');; I3 t9 u6 V( t3 y
    
41. subplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');
    3 F4 l4 `: {# Z( @- X% x
42. legend('网络输出货运量','实际货运量');
    ) a% \; N* c' v  T
43. xlabel('年份');ylabel('货运量、万吨');* F- O% o5 R( D
    
44. title('运用工具箱货运量学习和测试对比图');
    8 H, S/ [( }. D) w7 T
45. ! [$ z7 e( N) s& p
    
46. %新数据仿真
    , a6 _0 a- y& B7 R  V
47. pnew=[73.39,75.55# H: f, ]4 o0 g/ L! j2 L
    
48.      3.9635,4.0975
    $ @8 j+ i+ _6 P: t# l0 J2 g
49.      0.9880,1.0268];
    & X* ^* n. U  y+ g) c* K; v
50. pnewn=mapminmax('apply',pnew,ps1);& ?! v' o2 ]) a* `
    
51. anewn=sim(net,pnewn);/ r& Z) a* A9 F4 r% c8 u
    
52. anew=mapminmax('reverse',anewn,ps2)

复制代码

T-Eric

且生 发表于 2014-7-21 18:22
忽略我上面那个，改动的地方在18，21，25行……求批

- D0 R9 o) ], M2 R0 C确实改善了很多，很是感谢。但效果还是不太理想，而且为何只学习了几十次就停了呢？即使我修改了目标精度。

且生

1. %BP网络训练9 E% I8 G6 P/ U0 m# Y! t
   
2. net=newff(p0,t0,6,{'tansig'},'traingd');
   7 ]$ y( t/ v9 n$ ]4 G
3. net.trainParam.show=1000;
   + Z9 A: u4 H; k8 H
4. net.trainParam.Lr=0.05;
   \" X+ \6 w. o. q4 Z
5. net.trainParam.epochs=2000;
   $ ~  X( ?4 j3 a1 t1 M
6. net.trainParam.goal=0.65*10^(-4);
   ) g1 w2 w7 B2 G, ~6 t# y
7. net=train(net,pn,tn);

复制代码

我只改了这里面的，
一、改成单隐含层的，6个节点
4 x8 ]: X1 z+ w8 G% M/ o  O, v/ x二、训练函数改成梯度下降BP算法 traingd
5 e5 _6 g( C" O: o& C三、迭代次数改成2000
7 F* C+ c, K' Q$ M上面的参数是自己试的，我也不知道为什么。
; D7 X/ J; @: v1 k由于这玩意儿比较不靠谱，楼主多运行几次就能找到拟合的比较好的网络。
6 _& K/ L2 O7 f5 d9 U/ ~0 k6 ?关键问题是有没有过拟合我也不知道，等大神来解答吧

且生

因为样本比较少，所以训练函数没有选会调节学习率的