You are on page 1of 5

29 2

Vol. 29 No. 2

Control

and

2014 2
Feb. 2014

Decision

: 1001-0920 (2014) 02-0241-05

DOI: 10.13195/j.kzyjc.2012.1523

Hopfield

1,2 , 1 , 1

(1. 1001242. 102442)

: Hopfield (DHNN) , , DHNN

. (MD) , DHNN
. , , .
, , , . ,
, .
: Hopfield
: TP273

: A

An improved learning algorithm for discrete Hopfield


LI Rong1,2 , QIAO Jun-fei1 , HAN Hong-gui1
(1. School of Electric Information and Control EngineeringBeijing University of TechnologyBeijing 100124China;
2. Department of Water Conservancy and Architectural EngineeringBeijing Vocational College of AgricultureBeijing
102442ChinaCorrespondentLI RongE-maillirong0602@sina.com)
Abstract: According to the weight design problem of discrete Hopfield neural network(DHNN), an improved learning
algorithm for weight design is proposed. On the basis of the dynamic analysis for DHNN, the learning algorithm is designed.
The orthogonal matrix is got by using the method of matrix decomposition(MD), and the orthogonal matrix is used to get
the weight matrix of DHNN directly. The weight matrix which is obtained by the learning algorithm can store information
well, so that the testing sample can converge to a stable point. The learning algorithm does not need block calculation. The
calculation steps, the amount of calculation and the number of iterations are reduced, so the operating speed of the network
is improved. Finally, the algorithm is applied to water quality evaluation to prove its effectiveness and feasibility.
Key words: discrete Hopfield neural networkweight matrixmatrix decompositionwater quality evaluation

, [7] DHNN ,

Hopfield (DHNN)[1-2]

. Somesh [8]

, ,

, Hebb

. DHNN

DHNN . ,

, [3-4] [5] [6]

Hebb [9] ,

,
DHNN ,

; Hebb

, , DHNN

, , .

, .

[10] ,

. [11] Hebb

, , DHNN

, DHNN

. DHNN

, , ,

: 2012-10-15: 2013-01-23.
: (61034008)(61203099)
(4122006)(20121103120020).
: (1981), , , (1968), , ,
, .

242

29

. Srivastava [12]

( + 1) =

[1 ( + 1), 2 ( + 1), , ( + 1)].

. [13] Hebb
, ,
DHNN ,

(4)

( + 1) :
( + 1) = ( ()),

; , ,

DHNN .

, .

(5)

, (MD)

DHNN ,

. , DHNN ,

, ,

DHNN ; ,

, ,

. ,

2.1

() = [1 (), 2 (), , ()] DHNN

1 DHNN

. , DHNN

DHNN ,

1) 0, ;

, .

2) () ,

, () .

. , DHNN 1 .
-1

DHNN ,

(
)
=
() ,

= ( + 1) (), = 1, 2, , .

-1

1) () = ( + 1), = 1, 2, , ,

x 1( k +1)

x 2( k +1)

x n( k +1)

(2) (6) < 0, 0; ( +


1) = (), () .

2) () = ( + 1), = 1, 2, , ,

z-1

(6)

=1

(2) (6) = 0, 0; ( +
1) = (), () , ()

DHNN

DHNN,

, DHNN

+1 1.

, , DHNN

() {+1, 1}, = 1, 2, , . DHNN

(), .

, ,

(), ()

DHNN .

() = [1 (), 2 (), , ()].

(1)

DHNN , 1
+ 1

)
(
() .
( + 1) =

2.2

DHNN ,

. DHNN
,

(2)

: 1) DHNN ;

2) . MD

=1

;
() =

1, 0;
1, < 0.

DHNN ,
(3)

+1 ( +1),
( + 1)

DHNN .
. 1 , 2 , ,
, ,
T
= [1T , 2T , ,
].

(7)

: Hopfield

,
DHNN . ,
, MD
, (7) ,
= T .

(8)

: , ; .
,
, .
Hebb , , ,

, 1 :

1 1 1 1 1

1 1 1 1 1

1 1 1 1 1

1 1 1 1 1

1 =
1 1 1 1 1 .

1 1 1 1 1

1
1
1
1
1

1 1 1 1 1
1

MD , DHNN
.
DHNN , ,
, [3] ,
= diag[1 , 2 , , ].

III

IV

7.5

BOD5 /(mg/l)

10

/(/l)

10

15

0.02

0.02

0.02

0.2

0.2
10 000

200

1 000

2 000

5 000

/(mg/l)

0.002

0.002

0.005

0.01

0.1

/(mg/l)

1.0

1.0

1.0

1.5

1.5

0.01

0.05

0.05

0.05

0.1

I 1

= T .

II

/(mg/l)

()/(mg/l)

= 2 , 2 [0, 1), = 2, 3, , .

COD/(mg/l)

1 = 1 , 1 [1, 10];

(11)

/(mg/l)

(9)

243

, 2 , I ,
(10)

I .

MD , Hebb
; ,
, ,
.

3
MD
,
.
,
.

, 5
3 , ,
.

[14-15]
, .
,
, DHNN
. ,
DHNN ,

, .
BOD5 COD
() 8

II

III

IV

1 5 1 ,
2 , 3 , 4 , 5 , =

, IIIIIIIVV , 1

[1T , 2T , 3T , 4T , 5T ]. DHNN = 5,

. DHNN 5 . 5

= 5, 1 = 5, 2 = 0.2. (8),

1 , 2 , 3 , 4 , 5 .

, . (9),

I , I

. , (10), DHNN

+1 , I 1 ,

.
.

244

29

10

/(mg/l)

8.9

5.0

3.5

6.2

9.7

10.4

7.5

4.9

6.9

4.0

BOD5 /(mg/l)

0.6

2.8

6.0

4.4

1.4

1.2

0.7

6.3

1.6

3.0

COD/(mg/l)

1.2

4.6

8.1

5.4

3.1

1.2

0.7

6.5

3.3

1.2

0.013 2

0.023

0.05

0.188

0.000 4

0.000 5

0.000 5

0.000 5

0.061 3

0.02

/(/l)

2 390

12 000

5 000

3 000

2 000

2 380

2 000

5 000

2 380

2 380

/(mg/l)

0.001

0.001

0.005

0.003

0.001

0.001

0.001

0.038

0.001

0.003

/(mg/l)

0.138

4.399

1.0

0.979

0.313

0.054

0.109

0.130

0.926

1.0

()/(mg/l)

0.002

0.002

0.002

0.002

0.002

0.002

0.002

0.002

0.002

0.04

/(mg/l)

, 2

. 10 , 1

, 3 . 3

5 2

, Hebb ,

. 1 ,

, 8 . 3

1 ,

4 , ,

, 100%,

MD ,
Hebb .
3

10

II

III

II

III

II

Hebb

II

III

II

II

II

II

III

II

III

II

4 .

4 1

,
1 ,

. Hebb ,

, 5 . ,

, DHNN

, , 5 ,

, .
4

10

5 , MD

10

10

10

10

10

Hebb

DHNN , .

, 5 .
5 : 3 ,
, Hebb ,
;
Hebb , . ,
DHNN ,
.

10

,
, Hebb , Hebb
, ;
, ,
,
, . ,

: Hopfield

245

10

0.564 534

0.571 130

0.573 629

0.571 633

0.570 762

0.562 918

0.572 051

0.577 752

0.568 531

0.566 575

Hebb

0.000 336

0.000 347

0.000 306

0.000 387

0.000 339

0.000 321

0.000 413

0.000 308

0.000 305

0.000 292

0.009 410

0.008 269

0.008 349

0.007 627

0.006 855

0.004 957

0.005 837

0.003 855

0.003 746

0.002 022

[7]

, , . Hopfield
[J]. , 2006, 21(7): 771-775.

DHNN
, .

(Wang Z J, Fan H, Yan C. Dynamics of Hopfield neural

, MD

networks with dynamic synapses[J]. Control and Decision,

, ,

2006, 21(7): 771-775.)

[8]

neural network with sub-optimal and random GA for

, :

pattern recalling of English characters[J]. Applied Soft

1) MD , ,
Hebb ,
;
2) ,
, ,
, , ,
DHNN .
(References)

Somesh Kumar, Manu Pratap Singh. Study of Hopfield

Computing, 2012, 12(8): 2593-2600.


[9]

Sompolinsky H. The theory of neural networks: The Hebb


rule and beyond[J]. Lecture Notes in Physics, 1987, 275:
485-527.

[10] Rogerio

Martins

Gomes,

Antonio

Padua

Braga,

Henrique Elias Borges. Information storage and retrieval


analysis of hierarchically coupled associative memories[J].
Information Sciences, 2012, 195: 175-189.
[11] Wei Yan, Chun-Hsien Chen, Wunching Chang. An

[1]

Hopfield J J. Neural networks and physical systems with


emergent collective computational abilities[J]. Proc of the
National Academy of Sciences, 1982, 79(8): 2554-2558.

[2]

Hopfield J J. Neurons with graded response have collective


computational properties like those of two-state neurons[J].
Proc of the National Academy of Sciences, 1984, 81(10):

[3]

3088-3092.
Zheng Pengsheng, Tang Wansheng, Zhang Jianxiong. A
simple method for designing efficient small-world neural
networks[J]. Neural Networks, 2010, 23(2): 155-159.

[4]

network[J]. Computers and Industrial Engineering, 2009,


56(4): 1617-1626.
[12] Vipin Srivastava, Parker D J, Edwards S F. The nervous
system might orthogonalize to discriminate[J]. J of
Theoretical Biology, 2008, 253(3): 514-517.
[13] Amr A Adly, Salwa K Abd-El-Hafiz. Efficient modeling
of vector hysteresis using a novel Hopfield neural network
implementation of Stoner-Wohlfarth-like operators[J]. J of

time-varying Hopfield network[J]. IEEE Trans on Neural

Advanced Research, 2013: 4(4): 403-409.


[14] Han Honggui, Chen Qili, Qiao Junfei. Research on

Gonzalo Pajares, Mara Guijarro, Angela Ribeiro. A

an online self-organizing radial basis function neural

Hopfield neural network for combining classifiers applied

network[J]. Neural Computer & Application, 2010, 19(5):

to textured images[J]. Neural Networks, 2010, 23(1): 144-

667-676.

153.
[6]

using a design knowledge hierarchy and Hopfield

Lee Donq-Liang. Pattern sequence recognition using a


Network, 2002, 13(2): 330-342.

[5]

investigation into sustainable product conceptualization

[15] Han Honggui, Qiao Junfei. A self-organizing fuzzy neural

Toru Aonishi, Yasunao Komatsu, Koji Kurata. Self-

network based on a growing-and-pruning algorithm[J].

consistent signal-to-noise analysis of Hopfield model with

IEEE Trans on Neural Network, 2010, 18(6): 1129-1143.

unit replacement[J]. Neural Networks, 2010, 23(10): 11801186.

()

You might also like