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！！
+ }: f) }0 Q7 Q1 a
* L7 L1 _" L+ |%原始数据输入
1 m7 b2 M- N' T6 q; p5 P: qclc
3 X/ V, m" o& s" c- \% {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];
1 t0 F# R  Y; D# Zsqjdcs=[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];
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,...
    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];
* Q' L+ M. ^0 W# x$ v$ kp=[sqrs;sqjdcs;sqglmj];
" c, s0 x8 l; o: G2 }' @t=[glkyl;glhyl];

4 s+ s9 z+ A8 U, U5 B3 l) C: a8 p  m%数据归一化
[pn,ps1]=mapminmax(p);
[tn,ps2]=mapminmax(t);
dx=[-1,1;-1,1;-1,1];
) v, N7 B8 S% }9 ?9 O0 c
%BP网络训练
net=newff(p,t,[3,7,2],{'tansig','tansig','purelin'},'traingdx');
2 D: G  C2 m0 G1 b! d, s4 znet.trainParam.show=1000;
net.trainParam.Lr=0.05;
net.trainParam.epochs=50000;
1 t9 D/ Z" X+ x" R- h1 N0 c0 ^net.trainParam.goal=0.65*10^(-3);
8 {& ?+ T2 O. S2 i8 onet=train(net,pn,tn);

%利用原数据对BP网络仿真
an=sim(net,pn);
2 j0 f( B8 h& ~. ?0 Wa=mapminmax('reverse',an,ps2);
5 o: t7 r, l" t, k, T, M; w
%仿真结果与原数据对比测试
x=1990:2009;
! J1 ?6 h  p- Fnewk=a(1,:);
2 t/ Z8 M. w" S0 A( ?6 Fnewh=a(2,:);
figure(2);
subplot(2,1,1);plot(x,newk,'r-o',x,glkyl,'b--+');
legend('网络输出客运量','实际客运量');
+ o" m9 {* c4 Q" axlabel('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');
subplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');
; d& W3 j6 U& g5 M+ j5 L. Llegend('网络输出货运量','实际货运量');
9 T$ F3 }- Q7 E( m+ K4 dxlabel('年份');ylabel('货运量、万吨');
title('运用工具箱货运量学习和测试对比图');
6 L- O/ o3 b& j4 a. ?
2 L. ?8 p- ~7 {) N%新数据仿真
2 b0 K+ I' g! |( w  d: @) d/ Z2 x5 |: ^pnew=[73.39,75.55
    3.9635,4.0975
9 ^8 d, j/ ^) ]8 Q1 w# Z# f/ p6 O    0.9880,1.0268];
# ]% Q# i9 `3 b- Q" `8 bpnewn=mapminmax('apply',pnew,ps1);
anewn=sim(net,pnewn);
9 w* x/ D& U0 g! i5 ^anew=mapminmax('reverse',anewn,ps2)

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
& N- u% X+ g" X/ ^+ I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：
3 h  g# J. u- H3 W1 pnet=newff(dx,[3,7,2]{'tansig','tansig','purelin'},'traingdx');
我修改为：
net=newff(p,t,[3,7,2],{'tansig','tansig','purelin'},'traingdx');
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：
1 |0 V. {. h$ F4 C[pn,minp,maxp,tn,mint,maxt]=premnmx(p,t);
4 f% W8 t9 j6 w; O) e5 o# w+ pa=postmnmx(an,mint,maxt);
6 p+ [4 J0 u; X& `8 Dpnewn=tramnmx(pnew,minp,maxp);
4 B  J2 U4 E. R! x' t: F' \anew=postmnmx(anewn,mint,maxt)
' E/ H( J- [& J6 O: Y) e% f* r  n. O1 u2 m我修改为：
+ ^  P! A! Z+ Y- L[pn,ps1]=mapminmax(p);[tn,ps2]=mapminmax(t);
% M4 v0 W0 K" Y; ^a=mapminmax('reverse',an,ps2);
pnewn=mapminmax('apply',pnew,ps1);
anew=mapminmax('reverse',anewn,ps2)

1 |# m/ v: ^0 c! ^% @, G原程序为：
%原始数据输入
, W8 |2 C  T0 V9 Z: Mclc
9 ~4 q, O% f) D' isqrs=[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,...
& e' Z- n  {8 R& g1 S    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 p2 Y$ ~' O& v! t    22598,25107,33442,36836,40548,42927,43462];
- p& h7 Q4 }) s$ J/ ?8 W; Gglhyl=[1237,1379,1385,1399,1663,1714,1834,4322,8132,8936,11099,11203,10524,11115,...
    13320,16762,18673,20724,20803,21804];
/ Z0 v, v  n8 o6 @- T+ M" g: ip=[sqrs;sqjdcs;sqglmj];
t=[glkyl;glhyl];
1 a* i6 m5 P2 w1 @5 P
0 I: Q) U+ D9 [4 |/ ~0 B) @; d$ z%数据归一化
# p" S# ]  z9 q/ Q' d) W9 c; |! C[pn,minp,maxp,tn,mint,maxt]=premnmx(p,t);
dx=[-1,1;-1,1;-1,1];

* Y+ J( Y9 j. {0 K% D3 M' p8 h%BP网络训练
2 i2 C( T5 F+ m9 Z$ pnet=newff(dx,[3,7,2],{'tansig','tansig','purelin'},'traingdx');
net.trainParam.show=1000;
net.trainParam.Lr=0.05;
net.trainParam.epochs=50000;
net.trainParam.goal=0.65*10^(-3);
net=train(net,pn,tn);
% R1 E# K1 P3 E3 [) h) O" V- w
& t7 {$ p# z0 |%利用原数据对BP网络仿真
an=sim(net,pn);
a=postmnmx(an,mint,maxt);
! l. O$ @# l2 b, `0 V
%仿真结果与原数据对比测试
5 z0 t6 I$ [. }$ a+ O/ ux=1990:2009;
- ^) c: G. f: k4 D+ Lnewk=a(1,;
0 D+ ^6 T6 e8 Z; ]. enewh=a(2,;
figure(2);
1 k( v7 j' K+ m8 }% |/ _subplot(2,1,1);plot(x,newk,'r-o',x,glkyl,'b--+');
' g: I& }/ U% P/ F7 Hlegend('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');
subplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');
4 `; ]! a5 G3 N3 l* Q* Alegend('网络输出货运量','实际货运量');
xlabel('年份');ylabel('货运量、万吨');
title('运用工具箱货运量学习和测试对比图');
+ d0 D8 c! Z. Q- n9 A( E
%新数据仿真
pnew=[73.39,75.55
2 q' r, j9 E* z; g    3.9635,4.0975
+ o1 ?8 D) L) {, v    0.9880,1.0268];
& n, N" T5 P. @1 [3 zpnewn=tramnmx(pnew,minp,maxp);
, ~' n9 ^- f. u9 ranewn=sim(net,pnewn);
) q, ~4 b8 W( F! F& I) `. F: I: |anew=postmnmx(anewn,mint,maxt)
9 H9 A4 D0 U! j+ P8 j% E$ z+ T
修改后程序为：
3 F' m$ G7 r( r* C& S/ A5 ~%原始数据输入
clc
! e4 r6 m/ C9 x9 Q3 Y7 A  V( lsqrs=[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,...
) I  E% H4 q5 D1 k    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,...
3 k' B! n  |5 E- S/ F    0.56,0.59,0.59,0.67,0.69,0.79];
( k1 t2 B0 `" P0 L5 [! U( U  Rglkyl=[5126,6217,7730,9145,10460,11387,12353,15750,18304,19836,21024,19490,20433,...
    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];
  P9 y" t$ d; Y  c$ yp=[sqrs;sqjdcs;sqglmj];
* ^0 I5 L9 H* Nt=[glkyl;glhyl];
, S( o1 H/ A$ t+ j9 ]% g' _
%数据归一化
[pn,ps1]=mapminmax(p);
& B/ d5 e1 z# ?5 u& N[tn,ps2]=mapminmax(t);

%BP网络训练
net=newff(p,t,[3,7,2],{'tansig','tansig','purelin'},'traingdx');
net.trainParam.show=1000;
net.trainParam.Lr=0.05;
net.trainParam.epochs=50000;
net.trainParam.goal=0.65*10^(-3);
net=train(net,pn,tn);

* _- |4 U3 O0 q%利用原数据对BP网络仿真
an=sim(net,pn);
a=mapminmax('reverse',an,ps2);

%仿真结果与原数据对比测试
9 `2 S5 l0 k  W! C  I3 x6 h" ~& ?x=1990:2009;
newk=a(1,;
7 N3 v% t9 ~( ~' [# fnewh=a(2,;
figure(2);
subplot(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');
! c9 L. z# O. m$ N1 {3 A0 j* Qsubplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');
legend('网络输出货运量','实际货运量');
. F6 {- T: s( i) A& {1 @* y$ vxlabel('年份');ylabel('货运量、万吨');
1 M4 u/ E% Y3 `  ~title('运用工具箱货运量学习和测试对比图');

9 [+ ~7 @1 U6 D+ Z  C%新数据仿真
pnew=[73.39,75.55
    3.9635,4.0975
( M2 D6 C. U! H$ c% O; @; [    0.9880,1.0268];
, A' m: J+ G# X3 ^/ s2 m  K! dpnewn=mapminmax('apply',pnew,ps1);
! g5 Z+ e* S9 l! r/ g3 _2 n# ianewn=sim(net,pnewn);
anew=mapminmax('reverse',anewn,ps2)
（修改的地方用颜色标记了）
) C9 d! |4 F, I) |( ~3 J2 J1 ?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
   7 }' `1 n# I6 |
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,...
   9 v1 ^, C4 f6 h9 w+ [
3.     41.93,44.59,47.30,52.89,55.73,56.76,59.17,60.63];
   ) G  ?' s8 s# T+ E$ P: O+ B
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,...& M. e6 x. ~) d( X- y
   
5.     2.5,2.6,2.7,2.85,2.95,3.10];) c+ `2 Q6 H8 L/ x& O* j* V9 e
   
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,...
   / f9 \* x/ ~. j! ^\" [/ R& w5 C; S
7.     0.56,0.59,0.59,0.67,0.69,0.79];
   # R. B$ y8 X7 i7 _
8. glkyl=[5126,6217,7730,9145,10460,11387,12353,15750,18304,19836,21024,19490,20433,...# Q& G* m9 c) U! c6 e2 O
   
9.      22598,25107,33442,36836,40548,42927,43462];. z& q) E2 X5 Y2 S# ^/ E
   
10. glhyl=[1237,1379,1385,1399,1663,1714,1834,4322,8132,8936,11099,11203,10524,11115,...
    9 r: C2 \* o8 F6 [+ w
11.      13320,16762,18673,20724,20803,21804];1 ^; W) z  O7 l6 P
    
12. p=[sqrs;sqjdcs;sqglmj];
    2 F2 p! ~& u; R/ ?  ~  D! E; E) {: \% ~
13. t=[glkyl;glhyl];8 y, J3 R4 [1 d
    
14. %数据归一化
    5 b7 N) [6 W7 |2 ~
15. [pn,ps1]=mapminmax(p);9 P9 q8 q0 I2 G
    
16. [tn,ps2]=mapminmax(t);; Y; `6 h+ u\" v. j: C; @' U) q
    
17. %dx=[-1,1;-1,1;-1,1];
    5 r9 x1 u0 m8 [3 o1 K\" E# u
18. [color=Red] p0=minmax(pn);t0=minmax(tn);[/color]
    2 Y0 \2 K' Y, @
19. 
    : V- v* V, v. F
20. %BP网络训练2 G$ ^& E$ J\" @; Y, ?4 A
    
21. [color=Red] net=newff(p0,t0,[3,7,2],{'tansig','tansig','purelin'},'traingdx'); [/color]
    - r+ M8 X& o( u$ @
22. net.trainParam.show=1000;
    ! \1 T6 e+ y1 q
23. net.trainParam.Lr=0.05;
    / w8 D\" u4 L. _\" {* [( i2 q
24. net.trainParam.epochs=50000;; p+ ?; x& c9 ~8 G2 [
    
25. [color=Red] net.trainParam.goal=0.65*10^(-5);[/color]% o+ j$ Z, B/ v% j- j
    
26. net=train(net,pn,tn);
    ! q2 V, X( E9 P5 {. l+ M& [
27. 
    ; L* A( z' p$ W9 a: \$ E) T
28. %利用原数据对BP网络仿真1 e4 Q; Y9 G. Q$ ~1 R2 ?
    
29. an=sim(net,pn);1 D/ k5 Y7 I- f; y2 ?- e
    
30. a=mapminmax('reverse',an,ps2);4 V\" j\" K8 g) y. k2 d% O' U8 h
    
31. 
    2 f- y* b( S. w+ t
32. %仿真结果与原数据对比测试 5 @  u( b: J' h( z/ m* ?5 }  A
    
33. x=1990:2009;( @) D$ _1 V0 A- C  s/ c4 d
    
34. newk=a(1,:);* J: i) \2 n$ ?* _
    
35. newh=a(2,:);4 X: U, B6 ~3 R; h4 c% S
    
36. figure(2);
    3 _+ W2 U7 N1 E$ d6 o$ \: c9 V
37. subplot(2,1,1);plot(x,newk,'r-o',x,glkyl,'b--+');2 S' l  ?) m8 v
    
38. legend('网络输出客运量','实际客运量');
    2 Q: j$ E2 u0 J! ~7 @\" X
39. xlabel('年份');ylabel('货运量、万人');
    ; p) k' i7 O1 Q! T$ n  e% M. u
40. title('运用工具箱客运量学习和测试对比图');
    6 R* Z/ n* R7 o# j; \: u1 G
41. subplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');: ?/ v1 t/ k$ O0 V* z0 F# t8 u
    
42. legend('网络输出货运量','实际货运量');
    ) @\" l$ c9 Q- g+ m$ P* R
43. xlabel('年份');ylabel('货运量、万吨');
      a& b0 I% j% K# y9 U6 z
44. title('运用工具箱货运量学习和测试对比图');; A3 b$ w9 l8 ^$ P8 D
    
45. 
    % s! M9 J% U4 {$ w7 c* }; p
46. %新数据仿真0 j; |) N' P3 J; _- `
    
47. pnew=[73.39,75.55$ A\" `, z8 h8 e5 q
    
48.      3.9635,4.0975
    : V4 m8 L3 d9 d) q$ B8 ?# x% ?
49.      0.9880,1.0268];
    % J( z0 `\" d3 i, o, y- d
50. pnewn=mapminmax('apply',pnew,ps1);
    $ q9 F; F% L$ }. H5 `
51. anewn=sim(net,pnewn);; J$ p* s3 ~( F  l1 I0 g
    
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
   * [. R! R. E1 F\" V
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,...
   2 h) r8 d- [- i/ l: y$ k
3.     41.93,44.59,47.30,52.89,55.73,56.76,59.17,60.63];
   3 f. o0 C6 ?- }- L$ c6 X
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,...& T9 o5 l/ J1 X: |' X; {4 D
   
5.     2.5,2.6,2.7,2.85,2.95,3.10];
   9 ]; c5 i$ P% P6 l) n; i
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,...  L& a: n5 ~) k
   
7.     0.56,0.59,0.59,0.67,0.69,0.79];
   + o& W/ e\" E5 L9 @8 M5 x
8. glkyl=[5126,6217,7730,9145,10460,11387,12353,15750,18304,19836,21024,19490,20433,...- M4 R8 ~( b: F) \* U( @' w: F
   
9.      22598,25107,33442,36836,40548,42927,43462];
   0 Q- W4 w- j! I: l$ U
10. glhyl=[1237,1379,1385,1399,1663,1714,1834,4322,8132,8936,11099,11203,10524,11115,...; J1 p/ j5 H) j3 U/ J8 W2 J! v\" Y
    
11.      13320,16762,18673,20724,20803,21804];. X8 H* c$ k: H
    
12. p=[sqrs;sqjdcs;sqglmj];
    & [* |4 O5 X- ~9 Y# c6 T1 L
13. t=[glkyl;glhyl];7 @  P- z$ a6 I2 Z5 K' g
    
14. %数据归一化
    $ y+ t, N! b4 x$ m
15. [pn,ps1]=mapminmax(p);
    ; Y, T4 ^* U/ h, j& y8 P& E
16. [tn,ps2]=mapminmax(t);2 G& ?  N- u) N0 q
    
17. %dx=[-1,1;-1,1;-1,1];- |1 S: d7 ~, _1 v$ ]
    
18. p0=minmax(pn);t0=minmax(tn);
    - f! L/ G+ \' P! A+ x+ M, `
19. * s. e8 D% T/ U9 S6 u( X\" _* w\" z
    
20. %BP网络训练
    2 F; {0 I( c. a5 |# M8 S. o
21. net=newff(p0,t0,[3,7,2],{'tansig','tansig','purelin'},'traingdx'); # p4 M( B' P6 C: v- v1 Y: y9 G: c
    
22. net.trainParam.show=1000;
    5 r  s& E# X% l0 C
23. net.trainParam.Lr=0.05;
    5 {  Y9 |% q$ N- r: s0 G4 I; }: A
24. net.trainParam.epochs=50000;
    6 U  @4 y; U\" @% g  Y
25. net.trainParam.goal=0.65*10^(-5);
    ( }, u- ]) ^. Z) a
26. net=train(net,pn,tn);
    # S% h$ w' ~2 i7 g0 {( H
27. 6 h: J4 c8 p* [& o, |
    
28. %利用原数据对BP网络仿真, g8 l8 ]( `2 u! J8 M' |
    
29. an=sim(net,pn);
    6 u4 o  T8 d9 T7 \
30. a=mapminmax('reverse',an,ps2);
    6 V6 k3 o+ f$ _
31. 
    ; r\" e: u; }  l\" Q# h
32. %仿真结果与原数据对比测试 ( N3 z: k3 |8 B9 j/ a, n1 ]. ~
    
33. x=1990:2009;3 {& l& o0 k/ B) \+ f
    
34. newk=a(1,:);2 ?( Z0 ^' U/ J$ ^3 ?8 x. q; ]
    
35. newh=a(2,:);
    9 i9 h0 c, a\" B7 D8 o
36. figure(2);
    0 P( G' j8 {/ l! Q8 k1 p7 S
37. subplot(2,1,1);plot(x,newk,'r-o',x,glkyl,'b--+');
    & m4 b3 k* |! ]; s\" ^% D
38. legend('网络输出客运量','实际客运量');
    / z5 Q& P0 r# {( O. v! g
39. xlabel('年份');ylabel('货运量、万人');
    & ?, A9 Z5 P2 R0 @8 J4 ^
40. title('运用工具箱客运量学习和测试对比图');
    7 ?6 w( u# a/ a5 y
41. subplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');
    + I  `, Y2 s: c. T( v! X* q4 h# x
42. legend('网络输出货运量','实际货运量');
    8 j$ U: x4 ]* y9 F6 E& {1 Z
43. xlabel('年份');ylabel('货运量、万吨');0 O# v\" Q9 f; x
    
44. title('运用工具箱货运量学习和测试对比图');
    ' y7 K& T! ^# O& v. G
45. 
    . D, h\" D/ s0 h6 [. v
46. %新数据仿真
    ) k' \: Q% e3 U* f/ n
47. pnew=[73.39,75.558 L! y0 {1 R8 S- z& D5 h$ }
    
48.      3.9635,4.0975
    : U8 b0 X0 z6 \
49.      0.9880,1.0268];
    ( C+ A; P/ m* J\" Q! ~7 l* s
50. pnewn=mapminmax('apply',pnew,ps1);+ F0 r' S+ f  k$ m+ w
    
51. anewn=sim(net,pnewn);
    ( Q+ a7 q- r0 B1 t* K+ z
52. anew=mapminmax('reverse',anewn,ps2)

复制代码

T-Eric

且生 发表于 2014-7-21 18:22
# S/ {/ z+ B6 H; l" h5 [4 T忽略我上面那个，改动的地方在18，21，25行……求批

  d& M% K6 l& l: U8 r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8 c$ I. ?. O& I2 ^$ [1 B) y* n
   
2. net=newff(p0,t0,6,{'tansig'},'traingd');
   \" ]6 z( N+ f' d: [. x
3. net.trainParam.show=1000;
   4 h/ Z0 e) r* E/ _5 i& _
4. net.trainParam.Lr=0.05;& Y7 j. |9 ?7 h
   
5. net.trainParam.epochs=2000;$ e6 O/ {! v4 k' Y6 c
   
6. net.trainParam.goal=0.65*10^(-4);
   / [. K4 b% r' B/ E% n; U
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
8 L4 @/ g8 x4 \% g7 L8 N+ _( A* q+ F二、训练函数改成梯度下降BP算法 traingd
: g5 A! B6 U) r三、迭代次数改成2000
  f  o8 z* d4 F$ q7 `4 |/ p, o4 Z上面的参数是自己试的，我也不知道为什么。
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。
3 |9 `# G0 a! R" N1 W" w4 D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