НКТЗI_ЛР2011
.pdf31
4.3 )-3 - *
4.3.14
0 , 0 , 8 -
.
4.3.2(& . .
8 o & & 0 0 .
) & , -
(0 ) &
I) -
( & (0),
, , &).
4.3.36 1 2 . 4.2.
4.3.4w H - - , 8- ) 6 6 ( . 4.1).
= 4.1 – x
. I ( ,
. 4.2) ) I J ) . 5
’ 8 o &.
= 4.2 – I
32
4.3.5>
8 o & 8 ) I
I 0 , 0 &,
.
4.3.66 .
4.3.74& .
4.4 /+ ,)+
4.4.1: .
4.4.2.
4.4.34 .
4.4.4: ).
4.4.56 ) () 00
) &).
4.4.66, J
(5 I. ), 8 8 0 -
.
4.5 78 + , * 3
1.5 : ’ 8 o &, & 0 ), ' ,
, ) 0 ), &
(& r )), .
2.%i i ' i 8i o &i .
3.> , )
0 ).
4.g H 8
o & '0
, ' , -
8? 6 .
5.h 8 I . o &-
, 8? 4 q . > -
.
33
6.g 8 . o & 0
0 1101**** ’ : a) 11011001, 01010010, 11001111, ) 11011001, 11011010, 01011111, ) 11011001, 11010010, 11011111? 6 .
7.g 8 o & - & ( ) ?
8.g 8 ) 0 ) - 8 o &?
9.4 I & r ) ) )
.
10.i 8i o &i .
11.g 8 o & I
, I 8 I -
?
12.8 ) I i&i .
13.8i o &i i i i .
14.> 8 8 o & - ) I ) .
15.> 8 8 o & )- I ) .
16.> 8 8 o & -- 8.
17.4 8 8 o & -
.
18.3 I & 8 o &.
19.= 8 ) 0 8
o &.
20.8 o & ’ -
.
34
5 A B B B A 5
KB B L B ’M: B L P
5.1 " *
6 8 x ) .
5.2 ) + " " *H + )+-/ +
8 x 0 I – ) - ). 8 H ) I ) I -
' , ' 8,
I . z H 0
8 & 0
) . 8 x 8 -
'H 0
) 0 0 H ' .
> 0 8 x -
) 0 Matlab . 5.1.
: 5.1 – 0 8 x
"! 4 /* |
3 " 3 |
|||
x = [1 0 1 1 1 0 1 1] |
H - |
|||
|
|
|
|
: 8 , 1 . |
y = [0 0 0 1 1 0 0 1] |
H . |
|||
net |
= |
newelm([0 |
1], |
( 0H 8 x, J H |
[5 |
1], {'tansig', |
5 I I 1 |
||
'logsig'}); |
|
) I, 0 & |
||
|
|
|
|
|
|
|
|
|
'tansig' 'logsig', . |
net |
= |
train(net, |
x, |
H 8 x. |
y); |
|
|
|
|
Y = sim(net, x); |
|
4 0H 8 net |
||
|
|
|
|
, J 0 |
|
|
|
|
x. |
35
1. > 8 x
&.
P = 1:0.05:10;
T = P*2 - 10;
Pseq = con2seq(P); Tseq = con2seq(T);
net = newelm ( P, T, 10, {'tansig', 'purelin'}) net.trainparam.goal = 0.001; net.trainparam.epochs = 2000;
net = train(net, Pseq, Tseq); Y = sim(net, Pseq);
P1 = [2 3 4 10 12 11 11.5 14]; P1seq=con2seq(P1);
sim(net, P1seq)
2. ( 8 x -
) ).
p1 = sin(1:20); % JE:^@a Z>:;9UIaU@a ?@ZU>C
t1 = ones(1,20); % >;JCIH<W> JE:^9Z9 Z>:;9UIaU9Z9 ?@ZU>C<, :IAU> 9W@U@lI
p2 = p1 * 2; % W:<Z@a Z>:;9UIaU@a ?@ZU>C
t2 = t1 * 2; % >;JCIH<W> JE:^9Z9 Z>:;9UIaU9Z9 ?@ZU>C<, :IAU> WA9;
p = [p1 p2 p1 p2]; % AEYH9: AD9W< t = [t1 t2 t1 t2]; % AEYH9: lICEa Pseq = con2seq(p);
Tseq = con2seq(t);
net = newelm (Pseq, Tseq, 10, {'tansig', 'purelin'}, 'traingdx');
net.trainparam.goal = 0.01; net.performfcn = 'sse'; net.trainparam.epochs = 1000;
[net, tr] = train(net, Pseq, Tseq); subplot(3, 1,1); semilogy(tr.epoch, tr.perf);
title('o9;@CY@ ;E:EBI'); xlabel('pJ9D@'); a = sim(net, Pseq);
time = 1:length(p); subplot(3, 1,2);
% figure(2);
36
plot(time, t, '--', time, cat(2, a{:}))
title(']EF<CGH>H HE?H<A>UUV ;E:EBI'); xlabel('T>?'); ylabel('-- - F>W>UI FU>TEUUV, - - A@DIW ;E:EBI');
p3 = p1 * 1.6;
t3 = t1 * 1.6;
p4 = p1 * 1.2;
t4 = t1 * 1.2;
pg = [p3 p4 p3 p4]; tg = [t3 t4 t3 t4]; pgseq = con2seq(pg); subplot(3, 1,3);
a = sim(net, pgseq);
plot(time, tg, '--', time, cat(2, a{ : } ) ) title(']EF<CGH>H HE?H<A>UUV ;E:EBI W9AICGU@a ?@ZU>C'); xlabel('T>?'); ylabel('---F>W>UI FU>TEUUV, -- A@DIW ;E:EBI');
5.3 )-3 - *
5.3.14
0 , 0 , 8 -
.
5.3.2(& . . 8
x & 0 0 .
) & , -
(0 ) &
I) -
( & (0),
, , &).
5.3.36 1 2 . 5.2.
5.3.4(, 8 x-
) ).
5.3.5>
8 x 8 ) I
I 0 , 0 &,
. : 8 8 x o &.
5.3.66 .
5.3.74& .
37
5.4 /+ ,)+
5.4.1: .
5.4.2.
5.4.34 .
5.4.4: ).
5.4.56 ) () 00
) &).
5.4.66, J
(5 I. ), 8 8 0 -
.
5.5 78 + , * 3
1. .8 x.
2.> 8 o &i x.
3.> 8 x - -
8 0.
4.> 8 x ) I .
5.. i 8i o &i x i Matlab.
6.6 8 ’ I
.
7.5 8 ) 8
’ .
8.I %.
9.> 8 ) I 8 -
’ .
10.= : 8 -
8 ’ .
38
6 A B B B A 6
KB B L B P ’M: B L ; B N
@ M
6.1 " *
6 8 ’ (8 @-
) ).
6.2 ) + " " *H + )+-/ +
' 0 - ), . h J
0 , 0
8 ), 0 -
. > 8 'H
, 0H .
4 8 8 , I
8 ) ) ) .
h & ) 0 -
8 ) . -
&, I 8 H -
I &.
6.2.1 " "2 > "
8 @ 8, J )- 0. z 8, 8, J 0
, 0 & 0 8 .
8 ) H0 8 0 -
H 8
. . -
0 (0 «8 H ») 8
0 0 ) 8 .
39
@, J ) 0, 0 )
8 , J , & - I ) I , 0 - 8 .
@, J ) 0, 0 ,
– 8 . 5 8 H .
> 0 8 @ (SOM)
) 0 Matlab . 6.1.
: 6.1 – 0 8 SOM
"! 4 /* |
3 " 3 |
|
|
H H 400 |
|
x = rand(1,400); |
) 0 ) - |
|
|
|
|
|
( 0H 8 net H |
|
net=newsom( |
) 0: - |
|
minmax(x), [2 5]); |
H & H0 minmax, I I |
|
|
– 2x5. |
|
|
. H 8 net (& H |
|
net=train(net, x); |
) , - |
|
|
||
|
x. |
|
|
4 0H 8 net |
|
y=sim(net,x); |
, ) |
|
I . , J - |
||
|
||
|
0 x. |
|
yc=vec2ind(y) |
> 0H |
|
& H . |
||
|
1. ( ) ) @
4*5, 1000
. > ~.( 10 -
.
40
P=rands(2,1000); plot(P(1,:), P(2,:), '+') net = newsom(P, [4 5]); net.trainparam.epochs=1000; net.trainparam.show=100; net=train(net,P);
% a=sim(net,P) hold on
plotsom(net.IW{1, 1}, net.layers{1}.distances);
Testdata = rands(2,10);
Testres = vec2ind(sim(net, Testdata)) plot(Testdata(1,:), Testdata(2,:), '*k')
6.2.2 " "2 LVQ
@ ) (Learning Vector Quantization, LVQ) – 0 I . @ 0-
I H & . 4-
, , 0 I H, 8
8 . h J H 8, -
0 I, 8, ) . ( -
0 ) I H , J H « »,
) . I )
, 8 LVQ 0
, J 0 .
8 LVQ H I @, -
H -
&, 8.
8 I, ) 0 0 @-
I. 6 I H , H
) .
> 0 8 LVQ - ) 0 Matlab . 6.2.