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！！
( Z; L, p- \+ e- `
' e/ g( ~3 {5 O3 H, S/ Y- w) T%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,...
. r% k! R2 y" D# |    2.5,2.6,2.7,2.85,2.95,3.10];
# _1 G( n) w' Esqglmj=[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,...
7 k0 p  }  R3 _3 {; j    0.56,0.59,0.59,0.67,0.69,0.79];
) x( M1 T; O! {8 \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,...
) v) v% [) }6 X7 O- j  L    13320,16762,18673,20724,20803,21804];
3 h. l$ d4 @& ^9 {  bp=[sqrs;sqjdcs;sqglmj];
t=[glkyl;glhyl];

: ^( B3 E, ]+ r2 R8 m9 S( ?%数据归一化
- ^; @3 v) y) S) i4 V# r) G[pn,ps1]=mapminmax(p);
( c# p0 u6 t" H3 |6 P# B[tn,ps2]=mapminmax(t);
7 m+ Z. d+ b* d, p7 Gdx=[-1,1;-1,1;-1,1];
0 r6 n3 x. p3 ?+ H
( k. u: Q' t" I6 z%BP网络训练
6 e3 {. z2 M5 k" y1 S' Snet=newff(p,t,[3,7,2],{'tansig','tansig','purelin'},'traingdx');
4 k* A' @" p4 `0 o- \7 c4 Wnet.trainParam.show=1000;
2 [- j2 a, `& z4 X/ jnet.trainParam.Lr=0.05;
net.trainParam.epochs=50000;
net.trainParam.goal=0.65*10^(-3);
net=train(net,pn,tn);

%利用原数据对BP网络仿真
' X0 |$ X) S7 O) t6 ?2 N0 Wan=sim(net,pn);
8 D) Q2 i9 l7 m4 ^& }a=mapminmax('reverse',an,ps2);
7 u" _7 u( A8 u- U9 \6 Q
%仿真结果与原数据对比测试
x=1990:2009;
) t  m3 J" W6 h, S; t2 ^5 M5 Enewk=a(1,:);
. Z* \8 `& }1 z$ T6 k: J, A; m0 inewh=a(2,:);
! \: v7 {# w2 R1 {8 D: ]figure(2);
  ?$ ^2 I8 ?1 s7 q( ]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');
$ @6 J, G/ g7 P5 r; wtitle('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--+');
legend('网络输出货运量','实际货运量');
2 `. |* u7 I- Sxlabel('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');
" B9 ^& n. J- ?
%新数据仿真
pnew=[73.39,75.55
( V. i1 b, l; u$ N* n    3.9635,4.0975
    0.9880,1.0268];
/ n% y! \, [' v7 g) [pnewn=mapminmax('apply',pnew,ps1);
8 B& J# m4 |9 s" ]: L, q- [anewn=sim(net,pnewn);
anew=mapminmax('reverse',anewn,ps2)

- ^* b; {& B* G

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
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');
. _' c1 B# S, a; x+ X我修改为：
3 _- c; o6 x+ q7 V4 Snet=newff(p,t,[3,7,2],{'tansig','tansig','purelin'},'traingdx');
8 e0 g; d7 J* H8 S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);
% m4 L3 N+ [6 J; H1 @8 Za=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);
a=mapminmax('reverse',an,ps2);
pnewn=mapminmax('apply',pnew,ps1);
anew=mapminmax('reverse',anewn,ps2)

原程序为：
%原始数据输入
, m- [+ @  w# t* }1 bclc
9 s/ k* j3 b9 {/ x& 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];
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,...
* l' k: S7 H; d- v, Y2 l4 w/ u    2.5,2.6,2.7,2.85,2.95,3.10];
' c# w) q% C& h' g# T+ M" jsqglmj=[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];
8 y  c9 }+ G# c3 R! ^3 ^glkyl=[5126,6217,7730,9145,10460,11387,12353,15750,18304,19836,21024,19490,20433,...
    22598,25107,33442,36836,40548,42927,43462];
) J. R# A3 L, x( Cglhyl=[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];

%数据归一化
[pn,minp,maxp,tn,mint,maxt]=premnmx(p,t);
dx=[-1,1;-1,1;-1,1];
/ d% o9 Q/ h6 w3 Y( D# S- r. Q
%BP网络训练
; ?. y; Z5 F. ^) R' w8 \6 Q5 O' Snet=newff(dx,[3,7,2],{'tansig','tansig','purelin'},'traingdx');
net.trainParam.show=1000;
, d# C6 ~1 V$ F- Dnet.trainParam.Lr=0.05;
net.trainParam.epochs=50000;
net.trainParam.goal=0.65*10^(-3);
3 C7 ~8 I$ c. qnet=train(net,pn,tn);

%利用原数据对BP网络仿真
( f0 Z2 U- l( e# R& p1 Wan=sim(net,pn);
a=postmnmx(an,mint,maxt);

3 I6 q5 E  B) e  |& k$ `+ z%仿真结果与原数据对比测试
, x* y5 Z+ P+ H. L5 Q, Xx=1990:2009;
newk=a(1,;
: b9 s" }! P: L4 Rnewh=a(2,;
4 X- O$ K# u/ Z5 afigure(2);
$ ]& |+ B0 D8 ~, r: H9 jsubplot(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');
! h0 ~5 r' R" M" z+ Etitle('运用工具箱客运量学习和测试对比图');
! W9 W& Y7 l. F( q2 zsubplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');
legend('网络输出货运量','实际货运量');
1 W4 t# i" Q  ?6 e1 S, i2 jxlabel('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');
4 h/ P* b$ l# s0 j
%新数据仿真
- `9 k2 P1 A+ I+ F4 r+ rpnew=[73.39,75.55
    3.9635,4.0975
    0.9880,1.0268];
pnewn=tramnmx(pnew,minp,maxp);
/ x, A3 c2 f# b- }9 P$ Manewn=sim(net,pnewn);
anew=postmnmx(anewn,mint,maxt)
( G( h" J1 N1 a6 U9 H% \0 J% V
修改后程序为：
: d  K5 X1 X4 [9 S0 l%原始数据输入
4 B3 ~0 u! k# I2 _% \* S" P6 a- v' Qclc
& I" r' |/ a; m# g( ^5 C( [+ ]( Tsqrs=[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 i% u+ |1 }. g! a    41.93,44.59,47.30,52.89,55.73,56.76,59.17,60.63];
- s, |% O! _' |0 asqjdcs=[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 I) p0 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];
; Q0 J* S: V7 Q$ t+ N1 jglkyl=[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,...
$ e/ D, D( j3 H9 Z* s- Z    13320,16762,18673,20724,20803,21804];
p=[sqrs;sqjdcs;sqglmj];
t=[glkyl;glhyl];

( h% `0 T5 k6 R%数据归一化
5 p6 I) E; D# _) |- _  i0 ?[pn,ps1]=mapminmax(p);
[tn,ps2]=mapminmax(t);
1 J# `( ^! b# U
%BP网络训练
# N' E& g& K0 z# `& k+ Tnet=newff(p,t,[3,7,2],{'tansig','tansig','purelin'},'traingdx');
net.trainParam.show=1000;
) O, u1 H: p) e) d5 b$ D; cnet.trainParam.Lr=0.05;
8 {, ^; C. W) G3 h0 W6 Vnet.trainParam.epochs=50000;
3 U! s' Y! C7 |* _) g6 t. R6 Cnet.trainParam.goal=0.65*10^(-3);
net=train(net,pn,tn);

%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
x=1990:2009;
4 c. R, x: w8 Znewk=a(1,;
newh=a(2,;
. d6 D, z- I: ^$ m- e7 vfigure(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');
& z' a+ g" ^9 x! Oxlabel('年份');ylabel('货运量、万人');
% ]0 g" Q/ H8 |title('运用工具箱客运量学习和测试对比图');
1 B  X; o) G" R0 B3 Fsubplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');
; w. g' C) U9 `- `+ l! H# ?legend('网络输出货运量','实际货运量');
9 S3 J/ z( U( V! C& i2 s) P& Bxlabel('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');

$ P5 E7 {# ~% a6 A3 I2 o6 o2 }. I%新数据仿真
pnew=[73.39,75.55
    3.9635,4.0975
! G9 q% t( K/ Y# {    0.9880,1.0268];
pnewn=mapminmax('apply',pnew,ps1);
anewn=sim(net,pnewn);
anew=mapminmax('reverse',anewn,ps2)
! x" v( y/ D5 Y% c1 [  s5 Z0 O# n9 S（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. clc2 R6 g$ D  g/ A. e4 W
   
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,...; {\" Q\" ?: [5 T; k4 ]/ l- |8 @3 {$ U
   
3.     41.93,44.59,47.30,52.89,55.73,56.76,59.17,60.63];+ \\" X6 w\" U\" g2 }- @$ d
   
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,...
   8 r% F3 I; I! g% x1 i* f# Z8 w
5.     2.5,2.6,2.7,2.85,2.95,3.10];. W6 k  X8 p3 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,...
   1 v2 m4 X! l/ j. j- O
7.     0.56,0.59,0.59,0.67,0.69,0.79];# N( m! ^$ ?5 |! m. p% `
   
8. glkyl=[5126,6217,7730,9145,10460,11387,12353,15750,18304,19836,21024,19490,20433,...\" }3 K- g  `- D+ C
   
9.      22598,25107,33442,36836,40548,42927,43462];
   , p& n+ u# e! G8 m: ?; m
10. glhyl=[1237,1379,1385,1399,1663,1714,1834,4322,8132,8936,11099,11203,10524,11115,...: w, R5 d: H7 M' J& ]- Q0 X' `
    
11.      13320,16762,18673,20724,20803,21804];& Q% D# |' h/ N4 m6 f
    
12. p=[sqrs;sqjdcs;sqglmj];
    1 e4 U/ R; |2 B) V9 @) o3 P
13. t=[glkyl;glhyl];
    - s6 B  E: x& D\" h6 x
14. %数据归一化
    # L; j# d4 V, }, ~% k
15. [pn,ps1]=mapminmax(p);7 g5 r; l' c  o+ p8 Z\" V
    
16. [tn,ps2]=mapminmax(t);
    8 t2 w1 d8 g  S- _* [, _
17. %dx=[-1,1;-1,1;-1,1];
    ( g\" H+ X\" l. ]4 c9 L- C* L6 [
18. [color=Red] p0=minmax(pn);t0=minmax(tn);[/color]
    8 D4 p$ Q, S' P& Y: G
19. ; m* ~/ J+ U( R2 U- J. g
    
20. %BP网络训练
    # Y; o+ b+ ?$ p
21. [color=Red] net=newff(p0,t0,[3,7,2],{'tansig','tansig','purelin'},'traingdx'); [/color]
    6 D/ o7 M2 Z5 D- |8 {( L
22. net.trainParam.show=1000;
    0 l$ p8 Q9 u) O% R! P6 J
23. net.trainParam.Lr=0.05;
    - ~  P% V5 G0 ]1 T- a' R
24. net.trainParam.epochs=50000;
    ) p\" g8 e7 v7 k, [
25. [color=Red] net.trainParam.goal=0.65*10^(-5);[/color]/ V; P) b; u8 f% Z; M+ e0 M3 z
    
26. net=train(net,pn,tn);! L6 ]/ [$ N6 b$ L# [1 m2 w
    
27. 
    \" N4 ?$ U9 s! I; w! e- ?; I
28. %利用原数据对BP网络仿真
    + N; q1 G0 k$ N\" H5 m, W! l. p& S: k
29. an=sim(net,pn);, w5 c+ h9 c# u4 G6 g( J
    
30. a=mapminmax('reverse',an,ps2);, R  H+ H: B7 @2 S! Q
    
31. 
    2 {1 c# r4 M) q3 o% H
32. %仿真结果与原数据对比测试
    4 z9 X1 Y0 \& L1 x& F7 L
33. x=1990:2009;% q  d! q0 j% ?\" I( |- B
    
34. newk=a(1,:);- H* x- I8 [$ a) }
    
35. newh=a(2,:);
    , t5 o9 a$ n* D' ~
36. figure(2);
    % n& H# y: |7 ]
37. subplot(2,1,1);plot(x,newk,'r-o',x,glkyl,'b--+');
    $ P& _4 W( _  f- o
38. legend('网络输出客运量','实际客运量');+ O, Y2 N0 X8 s0 F! y! G8 O' B
    
39. xlabel('年份');ylabel('货运量、万人');
    + s: M3 ]8 V) M; v% U
40. title('运用工具箱客运量学习和测试对比图');
    5 ^$ x2 D, c% P- ?. A
41. subplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');
    2 {( }4 |3 \9 x& L# C  l7 y  s
42. legend('网络输出货运量','实际货运量');% e) j- d3 Z: U* }
    
43. xlabel('年份');ylabel('货运量、万吨');
    . R! S% O9 j. t: o- q\" C
44. title('运用工具箱货运量学习和测试对比图');
    * [# B. X, N/ B1 m& r. E, I: T1 Q\" ^
45. : l1 n. i2 E+ E0 }. T) _\" u
    
46. %新数据仿真
      X% e7 Q8 q+ g\" r
47. pnew=[73.39,75.55+ M  A0 g: M1 f3 J- A
    
48.      3.9635,4.0975& E6 p2 T7 \+ ?9 A/ D; Y\" Q
    
49.      0.9880,1.0268];
    0 N\" u9 J% Q* @7 W/ I. l
50. pnewn=mapminmax('apply',pnew,ps1);; D9 A2 K8 s5 Z6 _
    
51. anewn=sim(net,pnewn);0 }% m5 U4 f1 s\" P3 ]6 `\" D- u
    
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+ t, H2 r' R, F
   
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 B5 q5 k* f+ |6 E* q
3.     41.93,44.59,47.30,52.89,55.73,56.76,59.17,60.63];
   / N0 j# ]6 j* Z# c8 Z5 Y9 v
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,...
   + |& ^( C4 v9 L+ n, t6 W* j
5.     2.5,2.6,2.7,2.85,2.95,3.10];! D& a' \; U, i5 ?\" b/ 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,...# O! J( O5 F' l
   
7.     0.56,0.59,0.59,0.67,0.69,0.79];- h& p, p/ V5 b% P
   
8. glkyl=[5126,6217,7730,9145,10460,11387,12353,15750,18304,19836,21024,19490,20433,.../ ^+ A. ?. i# |  }2 v/ }  c& f
   
9.      22598,25107,33442,36836,40548,42927,43462];
   ; r6 N  s/ D; G0 {& e( S
10. glhyl=[1237,1379,1385,1399,1663,1714,1834,4322,8132,8936,11099,11203,10524,11115,...
    1 `; m! _! V1 g
11.      13320,16762,18673,20724,20803,21804];! `4 Q7 e; V/ B4 O: ?/ h+ _3 M/ F
    
12. p=[sqrs;sqjdcs;sqglmj];
    ) A# k9 u) d% x' |( J4 B
13. t=[glkyl;glhyl];
    * s3 U/ V7 @+ J, A3 m
14. %数据归一化6 M& Y  R2 a1 a( i8 L$ n\" {7 z' M% x
    
15. [pn,ps1]=mapminmax(p);# G  j; U; [$ H0 X' W
    
16. [tn,ps2]=mapminmax(t);
      k+ ]/ ]% r0 D$ c: [$ z! u
17. %dx=[-1,1;-1,1;-1,1];; B% A3 p6 ]8 k2 ]  Z+ T
    
18. p0=minmax(pn);t0=minmax(tn);\" u6 ~% C' o8 n  U4 d
    
19. 0 l! g% N. A) B& [7 `$ z
    
20. %BP网络训练1 _! A7 x1 c5 |* b: q3 q
    
21. net=newff(p0,t0,[3,7,2],{'tansig','tansig','purelin'},'traingdx'); % k2 X. M5 e% Z; U3 p
    
22. net.trainParam.show=1000;$ I9 s% ?/ O/ u/ y: D7 h\" ]; l
    
23. net.trainParam.Lr=0.05;\" ]# v7 P8 ?% X8 Y7 ~\" ]3 _
    
24. net.trainParam.epochs=50000;
    5 v  W& K' r7 q' G' C2 g5 o1 j- X$ w
25. net.trainParam.goal=0.65*10^(-5);
    \" p2 ~9 z  `9 I! {+ Y
26. net=train(net,pn,tn);5 J; U- _# ?; |& a6 z, ~
    
27. 
    7 n# g$ x* ~' [
28. %利用原数据对BP网络仿真- H' L% y& R  e  z& Z1 N
    
29. an=sim(net,pn);6 N3 K, R/ y4 l* }$ x4 h/ a
    
30. a=mapminmax('reverse',an,ps2);
    ; p  R& Q5 v% Z+ Q/ J0 \! @. e2 a
31. : H8 N3 L7 J% o) z0 r& _2 l! G
    
32. %仿真结果与原数据对比测试 7 g/ L: D! e; z\" {
    
33. x=1990:2009;& v2 \  e/ Y5 c  c\" p: W+ z
    
34. newk=a(1,:);
    ; ~9 E8 z1 x! r; Y- @% Z, G9 L- x3 v
35. newh=a(2,:);0 s  ]2 y% d) t  y8 N, _( N, y
    
36. figure(2);2 T8 E6 Q7 l\" L* }+ C0 Z
    
37. subplot(2,1,1);plot(x,newk,'r-o',x,glkyl,'b--+');\" |: Q. C5 l3 D6 a( G' w( g
    
38. legend('网络输出客运量','实际客运量');, s8 `( F6 |( G! a8 P\" g
    
39. xlabel('年份');ylabel('货运量、万人');
    2 Y: s6 T) L6 D8 ?
40. title('运用工具箱客运量学习和测试对比图');8 g! Z9 V* s( e/ D$ f3 j0 L
    
41. subplot(2,1,2);plot(x,newh,'r-o',x,glhyl,'b--+');
    ! w$ U1 x% o1 v4 [2 V
42. legend('网络输出货运量','实际货运量');! k$ U% Y2 [  z3 |, X4 |
    
43. xlabel('年份');ylabel('货运量、万吨');
    8 W$ w2 K: @$ ~/ _0 m
44. title('运用工具箱货运量学习和测试对比图');
      ]% p5 e+ M. c: ?# p. W0 z0 @
45. . ~; P& y\" M  s: K! `- p
    
46. %新数据仿真
    % L0 n. Z; @1 ]( `. s/ c
47. pnew=[73.39,75.55
    7 G% C7 ?3 u: G8 L$ Z: ?
48.      3.9635,4.0975+ @9 p! h0 \; h+ d; i2 X1 a4 A) b
    
49.      0.9880,1.0268];
    . s) @7 M# q, [9 ~
50. pnewn=mapminmax('apply',pnew,ps1);1 B+ q/ W\" {& c1 ^2 O& ^4 u; J
    
51. anewn=sim(net,pnewn);# J4 e( s& p+ _2 A* F+ ^  B
    
52. anew=mapminmax('reverse',anewn,ps2)

复制代码

T-Eric

且生 发表于 2014-7-21 18:22
- j: b$ f* N9 y% o4 ]3 \忽略我上面那个，改动的地方在18，21，25行……求批

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. Q6 `; G& S% Q1 n7 |1 s
   
2. net=newff(p0,t0,6,{'tansig'},'traingd');
   $ U  }; g- a. h- G1 e! Q
3. net.trainParam.show=1000;$ R$ [% y$ T+ Z9 P0 L: V
   
4. net.trainParam.Lr=0.05;
   . M% ?/ F% A5 d9 g7 E
5. net.trainParam.epochs=2000;2 i3 M+ i/ ]/ Q% D
   
6. net.trainParam.goal=0.65*10^(-4);3 i, G2 [/ g5 A3 ]+ D$ ]6 F
   
7. net=train(net,pn,tn);

复制代码

我只改了这里面的，
7 E' X2 Y, I  @7 Z, l一、改成单隐含层的，6个节点
8 T5 W! X3 J- O, l" `. W! l二、训练函数改成梯度下降BP算法 traingd
! ~2 N- g/ \" J+ J2 a三、迭代次数改成2000
. J; g; D' Z* g) B; l# h' X  L0 h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。
* \7 o1 c! G$ E& g) J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