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
clc
+ E) Y2 ?, D9 C4 x* F) I* ^- H+ _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 p  p( E8 \2 B3 [    2.5,2.6,2.7,2.85,2.95,3.10];
/ \+ f% ]$ ?- Y! Vsqglmj=[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,...
$ k2 ?2 j# G$ a( g- S# `    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,...
/ B& W# g6 a+ K6 I( l$ B    13320,16762,18673,20724,20803,21804];
) b. h+ Z8 d+ c$ a4 tp=[sqrs;sqjdcs;sqglmj];
t=[glkyl;glhyl];

%数据归一化
[pn,ps1]=mapminmax(p);
8 B$ X- H' D1 |) r[tn,ps2]=mapminmax(t);
& W3 [3 F! Z/ g1 R1 d) i7 y- t+ X) Pdx=[-1,1;-1,1;-1,1];

" e9 F9 l, K# s& e. H$ I/ i5 ~%BP网络训练
+ o; b( L, o3 t! ]  s% Snet=newff(p,t,[3,7,2],{'tansig','tansig','purelin'},'traingdx');
2 `' Q( p  D" X6 z2 R$ m: Pnet.trainParam.show=1000;
2 G6 H2 O4 e* f8 W2 G# M9 tnet.trainParam.Lr=0.05;
  r4 K: ]* K' a1 D0 t9 Wnet.trainParam.epochs=50000;
net.trainParam.goal=0.65*10^(-3);
net=train(net,pn,tn);
! |: I' F" S" i" W
3 R' w# m0 H' f$ |& \%利用原数据对BP网络仿真
1 h, Q5 Z/ P/ P- U# o2 I* C4 fan=sim(net,pn);
9 K' {- c  G0 ~* G$ V7 d+ Qa=mapminmax('reverse',an,ps2);

%仿真结果与原数据对比测试
' X! T$ ?+ q, Y' J5 f/ Ox=1990:2009;
/ c# j5 Y# z6 j  l8 Znewk=a(1,:);
; h; x5 m: n. X/ w% N' inewh=a(2,:);
figure(2);
; [7 F, f6 f9 z  Gsubplot(2,1,1);plot(x,newk,'r-o',x,glkyl,'b--+');
6 _/ I3 C5 f4 {. c4 O* u6 L" Qlegend('网络输出客运量','实际客运量');
% M' o1 I6 E2 L  V" nxlabel('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--+');
$ z. X4 T. S  r- A1 l, Y, dlegend('网络输出货运量','实际货运量');
xlabel('年份');ylabel('货运量、万吨');
( A- O9 R# k3 z7 |title('运用工具箱货运量学习和测试对比图');

%新数据仿真
pnew=[73.39,75.55
    3.9635,4.0975
    0.9880,1.0268];
: [* H( }0 p5 `% }# s  |pnewn=mapminmax('apply',pnew,ps1);
" Z, v3 V( x% r( j7 Ganewn=sim(net,pnewn);
- a( o3 J) h# h4 g" Yanew=mapminmax('reverse',anewn,ps2)
, |" u) I! b( |/ R3 x
/ n. }5 n  c+ l. i% d

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
! H( h& q6 Z- L$ s: B2 i; T1 E% u8 M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 ...

: d7 o- Z' k+ p& |因为版本更新，Matlab中的Newff命令用法有所改变，原命令为：
/ L9 o- l7 V* ]5 [; gnet=newff(dx,[3,7,2]{'tansig','tansig','purelin'},'traingdx');
我修改为：
9 i, ?/ ?# O/ T9 w" c6 u0 a& znet=newff(p,t,[3,7,2],{'tansig','tansig','purelin'},'traingdx');
5 u. Y3 u' n& ^0 ~2 [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);
0 ]9 S' {9 P1 i+ j( ga=postmnmx(an,mint,maxt);
4 I% B) U6 s" i( G7 j5 vpnewn=tramnmx(pnew,minp,maxp);
8 C; p* ^: q& Z0 E6 X. Fanew=postmnmx(anewn,mint,maxt)
4 j; _2 z( b' ]我修改为：
[pn,ps1]=mapminmax(p);[tn,ps2]=mapminmax(t);
2 x9 s- M( P, G% Ua=mapminmax('reverse',an,ps2);
pnewn=mapminmax('apply',pnew,ps1);
% _1 A% Z  b( S. W% S+ f2 P: O3 Yanew=mapminmax('reverse',anewn,ps2)
* b* C8 o1 V' D/ k6 G; ?; }0 E
原程序为：
- i; C# F3 M4 o% ^1 s9 `%原始数据输入
3 O# K# W2 s5 \7 n* h; O2 yclc
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,...
& W# V& }0 d2 O+ q    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];
9 K- P. f$ K+ z+ N0 Wglkyl=[5126,6217,7730,9145,10460,11387,12353,15750,18304,19836,21024,19490,20433,...
    22598,25107,33442,36836,40548,42927,43462];
7 K: J4 b2 }( j8 ]4 {glhyl=[1237,1379,1385,1399,1663,1714,1834,4322,8132,8936,11099,11203,10524,11115,...
3 m4 m) q7 t, }( T    13320,16762,18673,20724,20803,21804];
p=[sqrs;sqjdcs;sqglmj];
+ A; t# B. t5 [/ l  rt=[glkyl;glhyl];

%数据归一化
[pn,minp,maxp,tn,mint,maxt]=premnmx(p,t);
dx=[-1,1;-1,1;-1,1];
. a: T; F6 _5 t& U1 z& G7 f0 a
%BP网络训练
net=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);
/ |% L" e1 B' i  l* I; Bnet=train(net,pn,tn);

%利用原数据对BP网络仿真
an=sim(net,pn);
a=postmnmx(an,mint,maxt);

%仿真结果与原数据对比测试
- O& u! t2 V' c& R7 c1 mx=1990:2009;
$ m' N# U/ D, T: e9 H2 R+ j8 \newk=a(1,;
newh=a(2,;
figure(2);
subplot(2,1,1);plot(x,newk,'r-o',x,glkyl,'b--+');
0 U+ p( s% @; V$ i0 Q# f9 f" N; F& }legend('网络输出客运量','实际客运量');
3 T6 S2 p0 |( e# g2 D4 `$ h6 p+ kxlabel('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');
& k: x. o! J1 ^. rsubplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');
legend('网络输出货运量','实际货运量');
$ k! h/ B. d0 Axlabel('年份');ylabel('货运量、万吨');
; m4 [& g( I* }3 K; Rtitle('运用工具箱货运量学习和测试对比图');
# I1 c( V; V( U3 r" G, p9 y2 C
%新数据仿真
* K8 t, F' g+ R) |" `$ Y! `pnew=[73.39,75.55
    3.9635,4.0975
  D& k% i" o  h3 M* b    0.9880,1.0268];
pnewn=tramnmx(pnew,minp,maxp);
anewn=sim(net,pnewn);
5 ^3 N( H# j1 Ranew=postmnmx(anewn,mint,maxt)

修改后程序为：
%原始数据输入
clc
! Z8 `. H- s1 L/ }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];
+ P2 t5 g9 Y8 U) isqjdcs=[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,...
! K2 w+ k4 L+ d0 W2 b8 r2 @" z    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];
' {1 ~6 r/ {) uglhyl=[1237,1379,1385,1399,1663,1714,1834,4322,8132,8936,11099,11203,10524,11115,...
- n2 |6 T4 S% v1 p    13320,16762,18673,20724,20803,21804];
p=[sqrs;sqjdcs;sqglmj];
t=[glkyl;glhyl];
& S3 F% k5 Q0 P
1 H: p5 A# @7 j3 j2 y%数据归一化
[pn,ps1]=mapminmax(p);
2 ^0 Z/ M. {' e[tn,ps2]=mapminmax(t);
+ ^: Y. E( K" F4 J
- b! }4 D. \7 L- K%BP网络训练
0 k+ p7 g! O; knet=newff(p,t,[3,7,2],{'tansig','tansig','purelin'},'traingdx');
net.trainParam.show=1000;
net.trainParam.Lr=0.05;
8 Z) a& o* @# v& x! z0 x& I4 ]net.trainParam.epochs=50000;
1 m/ Q$ B3 H& `net.trainParam.goal=0.65*10^(-3);
net=train(net,pn,tn);

( e$ `3 E; S# x# z%利用原数据对BP网络仿真
( p0 \: R' Z* Y4 Xan=sim(net,pn);
a=mapminmax('reverse',an,ps2);

%仿真结果与原数据对比测试
x=1990:2009;
4 N7 l; F; \$ y6 @7 anewk=a(1,;
newh=a(2,;
2 y( `( u* o# ^$ `: ]2 Ofigure(2);
! u/ Z- W- c3 Zsubplot(2,1,1);plot(x,newk,'r-o',x,glkyl,'b--+');
6 t& a. `) X- i) y3 T- x0 Y* dlegend('网络输出客运量','实际客运量');
8 P! w3 @# l# ^( o. D7 _xlabel('年份');ylabel('货运量、万人');
& L& L( L  M  W+ y- \, ^( h0 f' Stitle('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--+');
- `+ d1 f+ d1 K; e. c- j2 C- vlegend('网络输出货运量','实际货运量');
) D3 l7 j6 O% N2 q! z8 T, Z  S# d- K1 O. `xlabel('年份');ylabel('货运量、万吨');
& H( P+ a! o1 \title('运用工具箱货运量学习和测试对比图');

3 w7 d; v4 _& n+ _2 r%新数据仿真
0 N, e4 z! C  o& d" @pnew=[73.39,75.55
; M, v0 H2 f$ Q) K( t    3.9635,4.0975
' i4 o. N; d  Q9 r- H    0.9880,1.0268];
pnewn=mapminmax('apply',pnew,ps1);
anewn=sim(net,pnewn);
9 a; t6 y* M8 B$ P, D! d7 C1 Danew=mapminmax('reverse',anewn,ps2)
（修改的地方用颜色标记了）
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' L\" W9 s( X. a* t6 a: G  y
   
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,...3 `! w/ q; f; P0 {, A* E+ X- r* [
   
3.     41.93,44.59,47.30,52.89,55.73,56.76,59.17,60.63];! h3 z6 Y5 _4 m# ?5 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,...- I, ^0 {# y5 p4 f! a
   
5.     2.5,2.6,2.7,2.85,2.95,3.10];4 X; }) B8 b8 X% p$ f7 A, Y
   
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,...
   1 |. q, I+ _- R* _: H\" U9 G
7.     0.56,0.59,0.59,0.67,0.69,0.79];8 M. W, f' a* @/ R6 `* o8 |
   
8. glkyl=[5126,6217,7730,9145,10460,11387,12353,15750,18304,19836,21024,19490,20433,...% G- g$ m3 B/ D7 c- x; F2 Y
   
9.      22598,25107,33442,36836,40548,42927,43462];0 `6 G! t) C+ g$ B  {
   
10. glhyl=[1237,1379,1385,1399,1663,1714,1834,4322,8132,8936,11099,11203,10524,11115,...
    ! i0 h: u\" Z2 ^% L  X( l( Z
11.      13320,16762,18673,20724,20803,21804];: f& U. W% R, s$ K
    
12. p=[sqrs;sqjdcs;sqglmj];+ C\" t\" K5 K% U% N! k' P: k
    
13. t=[glkyl;glhyl];
    6 m$ \/ p. h, k3 ]$ Z0 C
14. %数据归一化
    - y# l0 _! @  j. T
15. [pn,ps1]=mapminmax(p);8 A% B$ D! u* ]
    
16. [tn,ps2]=mapminmax(t);
    ) L: D3 M, |. s& s0 h- n
17. %dx=[-1,1;-1,1;-1,1];+ j, d( U) s: E( n' ~2 H* V. [$ Q7 T
    
18. [color=Red] p0=minmax(pn);t0=minmax(tn);[/color]9 H9 U; K3 P. @0 U, ^, Y1 m0 u9 I- X
    
19. 9 b3 p5 C' s2 ^0 L- A
    
20. %BP网络训练
    & A- [5 a. U$ O  I
21. [color=Red] net=newff(p0,t0,[3,7,2],{'tansig','tansig','purelin'},'traingdx'); [/color]
    \" [! s! {4 e  X; e
22. net.trainParam.show=1000;
    # k6 G2 ^/ o- a! X3 P
23. net.trainParam.Lr=0.05;
    . e9 {# {! A) F\" I5 D5 O
24. net.trainParam.epochs=50000;% K& _3 \& A) |9 N& c. l
    
25. [color=Red] net.trainParam.goal=0.65*10^(-5);[/color]
    0 U7 Y; U: @+ w* D  z5 Y\" f1 W8 Q
26. net=train(net,pn,tn);1 a, N6 u  u% ~8 B# L0 o0 w. D0 T) o
    
27. 6 o* S; c3 k  d3 F' b9 @% r2 R
    
28. %利用原数据对BP网络仿真0 W& V* o: |5 i8 g8 P- K6 I( z; i
    
29. an=sim(net,pn);9 u6 J0 k: D& {  p: p
    
30. a=mapminmax('reverse',an,ps2);
    ( J! s, ~' l% p  O* S7 e
31. 
    2 L+ U6 S8 P\" V5 I5 j/ p# ]
32. %仿真结果与原数据对比测试
    1 x  b2 |  |) r; X
33. x=1990:2009;) T' x/ _) [7 m1 s$ Z8 o% h
    
34. newk=a(1,:);
    ; D% k' a+ P, ]8 B
35. newh=a(2,:);- A3 ?/ a. }: m  H/ {# I2 U
    
36. figure(2);
    * r( p$ x\" L1 a# I+ Q5 O7 R
37. subplot(2,1,1);plot(x,newk,'r-o',x,glkyl,'b--+');
    \" J; Q# g5 t. S- W( L: B( q& F  u
38. legend('网络输出客运量','实际客运量');
      k9 B% j8 M, V
39. xlabel('年份');ylabel('货运量、万人');# k* A( x2 x+ P, Y0 n( w
    
40. title('运用工具箱客运量学习和测试对比图');. d% s9 }* D( [4 a0 l* |& E
    
41. subplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');, `& N9 f3 ~0 K: {3 u$ ~
    
42. legend('网络输出货运量','实际货运量');
    ' E+ f+ \7 Y6 l& |. d% ~. l) d) y/ G
43. xlabel('年份');ylabel('货运量、万吨');' o# B2 ^2 T+ C9 S
    
44. title('运用工具箱货运量学习和测试对比图');- f6 A' `- J1 A' k
    
45. 
      n. ~. E$ \7 P
46. %新数据仿真* q\" `8 W6 s  d. ~
    
47. pnew=[73.39,75.55# a! [. I0 h3 m7 A
    
48.      3.9635,4.0975
    # ?) U$ ~$ b0 }- {
49.      0.9880,1.0268];
    6 k  W3 N; t\" G1 G8 l9 L9 @. R% o7 j
50. pnewn=mapminmax('apply',pnew,ps1);9 V7 V\" l& c+ {3 Q6 Z% A/ r
    
51. anewn=sim(net,pnewn);  o+ U4 O$ h% F
    
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. clc5 _8 G8 }2 [' E- T& m) u8 o
   
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,...
   * B. `$ F( v) |) {\" W# Y  M
3.     41.93,44.59,47.30,52.89,55.73,56.76,59.17,60.63];, G# R! V' l9 J9 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,...
   ; \4 h0 Y( J; r: s! Y# b3 c
5.     2.5,2.6,2.7,2.85,2.95,3.10];; w4 _4 _1 v4 Z* W) o* F+ x
   
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,...
   8 ]1 N0 v/ s4 O1 r, T9 i2 H$ |
7.     0.56,0.59,0.59,0.67,0.69,0.79];. L5 Y/ f& G$ b3 l# L; I
   
8. glkyl=[5126,6217,7730,9145,10460,11387,12353,15750,18304,19836,21024,19490,20433,...8 y. H4 I7 U0 o1 O$ R- N( G0 q
   
9.      22598,25107,33442,36836,40548,42927,43462];
   5 N/ J# J$ K5 d9 a
10. glhyl=[1237,1379,1385,1399,1663,1714,1834,4322,8132,8936,11099,11203,10524,11115,...# h7 k: @5 h1 H0 t& b* F% W
    
11.      13320,16762,18673,20724,20803,21804];
    ( G1 B6 ^, q& Q
12. p=[sqrs;sqjdcs;sqglmj];
    $ M5 ]) k- Z  P3 ]- y; h
13. t=[glkyl;glhyl];8 l; i- c6 _; W9 S
    
14. %数据归一化. v4 }2 ~+ K9 p
    
15. [pn,ps1]=mapminmax(p);
    . U3 _1 w! i' m+ w2 s, h) l7 U9 a
16. [tn,ps2]=mapminmax(t);
    4 u1 [* f8 X* y) z/ E& Z9 d2 w
17. %dx=[-1,1;-1,1;-1,1];
    ; [, _+ u7 N& u, Q
18. p0=minmax(pn);t0=minmax(tn);
    # {3 ~\" l( z. w! ^7 C+ @
19. 
    ( v. D( F; y, t\" w+ y+ p
20. %BP网络训练
    2 t5 `! O) O9 l\" a2 V6 K
21. net=newff(p0,t0,[3,7,2],{'tansig','tansig','purelin'},'traingdx'); $ b6 v$ c' ^: X  k% ], J
    
22. net.trainParam.show=1000;5 k1 u  r8 O4 x  M# Q2 c
    
23. net.trainParam.Lr=0.05;
    # {  H: r\" X: c' k! M$ n
24. net.trainParam.epochs=50000;9 T% V- E1 E, t# F9 s5 o1 p
    
25. net.trainParam.goal=0.65*10^(-5);+ i( y' j3 z0 C3 b3 h) b  e& K
    
26. net=train(net,pn,tn);
    5 M$ r' _8 y% K6 p: s  i0 ^) R. ]
27. 7 n* m, U! S, S& w1 ]
    
28. %利用原数据对BP网络仿真/ q4 L  S, q+ f1 k! ^\" Q
    
29. an=sim(net,pn);$ N% m6 d: @8 P: |
    
30. a=mapminmax('reverse',an,ps2);6 c6 @' V7 l! ?
    
31. 
    + g$ I  L4 K! o* f5 M; V
32. %仿真结果与原数据对比测试
    # _/ s$ M4 q' T
33. x=1990:2009;
    % p7 ~1 p! _8 _* I4 l
34. newk=a(1,:);. h, v: H! @, ]3 R
    
35. newh=a(2,:);
    & f1 G: \* R9 R6 `
36. figure(2);* A8 `\" r5 j4 s0 V& L
    
37. subplot(2,1,1);plot(x,newk,'r-o',x,glkyl,'b--+');3 M/ h3 [/ ?7 s- {
    
38. legend('网络输出客运量','实际客运量');
    ) a1 I. c: A% I\" Y9 d% v% ^
39. xlabel('年份');ylabel('货运量、万人');5 S2 @  k/ u6 r# S% j' W0 g: }# T
    
40. title('运用工具箱客运量学习和测试对比图');+ Q2 j7 q$ B; b& m( t8 p; _
    
41. subplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');
    : h  s% c5 P: S. E6 F- N
42. legend('网络输出货运量','实际货运量');  C- `\" [* e6 a/ \
    
43. xlabel('年份');ylabel('货运量、万吨');% |! t# j0 \; @
    
44. title('运用工具箱货运量学习和测试对比图');' W5 I% J+ k\" H  H, D
    
45. ' U, @3 k/ M9 _# c% f
    
46. %新数据仿真, ?: F2 C* w: ?! e) T: a* g
    
47. pnew=[73.39,75.555 D) S. Y9 P( _$ B$ f. F
    
48.      3.9635,4.0975
    ! u( F6 M5 C5 f# J$ M
49.      0.9880,1.0268];: y/ D! Y. j/ n3 m' j
    
50. pnewn=mapminmax('apply',pnew,ps1);5 A! B- j4 }& I) p% v( E) d0 Q
    
51. anewn=sim(net,pnewn);
    \" O9 i  |4 U  \
52. anew=mapminmax('reverse',anewn,ps2)

复制代码

T-Eric

且生 发表于 2014-7-21 18:22
3 P% V/ L( K/ |* Q) q5 E9 |7 \忽略我上面那个，改动的地方在18，21，25行……求批

$ v9 n9 ?! D. h  r' A+ T+ u8 i) X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: }/ [, [; h* ]& Q7 m: x3 [8 L
   
2. net=newff(p0,t0,6,{'tansig'},'traingd'); 9 l4 {9 m' {& T\" `1 R: `& |& F
   
3. net.trainParam.show=1000;/ ^6 i8 k5 e  T9 N
   
4. net.trainParam.Lr=0.05;
   ) _& ]; d+ r$ I. }1 g' N' o  c% j% @
5. net.trainParam.epochs=2000;  |. T: `5 ~( w0 B
   
6. net.trainParam.goal=0.65*10^(-4);# A% Q5 S& A# M! j8 N  K& m5 C' `
   
7. net=train(net,pn,tn);

复制代码

我只改了这里面的，
  h, Z* J! [  J5 ~一、改成单隐含层的，6个节点
; C; E2 x1 _1 \+ _二、训练函数改成梯度下降BP算法 traingd
2 n1 `( {2 U  @* w三、迭代次数改成2000
3 ]: w  X$ p( @3 @2 S# Y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。
/ o/ M) ?! g" h' Q* B( W7 s2 z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