Proceedings of the 2006 IEEE

International Conference on Information Acquisition

August 20 - 23, 2006, Weihai, Shandong, China

An AND-OR Fuzzy Neural Network and Ship Application

Jianghua Sui Guang Ren
Marine Engineering College Marine Engineering College
Dalian Maritime University Dalian Maritime University
Dalian, Liaoning Province, 116026, China Dalian, Liaoning Province, 116026, China
suijhg tom.com renggdlmu.edu.cn
Abstract - A novel multilayer feed-forward AND-OR fuzzy equivalence is proved between the architecture of AND-OR
neural network (AND-OR FNN) and a piecewise optimization FNN and the fuzzy weighted Mamdani inference in order to
approach are proposed in this paper. The equivalent is proved utilize the AND-OR FNN to optimize fuzzy rules. The
between the architecture of AND-OR FNN and fuzzy weighted piecewise optimization of AND-OR FNN consists of two
Mamdani inference. The main superiority is shown in not only phases, first the blueprint of network is reduced by GA and
reducing the input space by special inner structure of neurons, P
but also auto-extracting the rule base by the structure PA; the second phase, the parameters are refinedby ACS (Ant
optimization of network. The optimization procedure consists of Colony System). Finally this approach is applied to design
two phases, first the blueprint of network is reduced by GA AND-OR FNN ship controller. Simulating results show the
(Genetic Algorithm) and PA (Pruning Algorithm); the second performance is much better than ordinary fuzzy controller.
phase, the parameters are refined by ACO (Ant Colony 1. Fuzzy NEURONS AND ToPOLOGY OF AND-OR FNN
Optimization). The AND-OR FNN ship controller system is
designed based on input-output data to validate this method. A. Logic-driven AND and ORfuzzy neurons
Simulated results demonstrate that the number of rule base is The AND and OR fuzzy neurons are two fundamental
decreased remarkably, the performance is much better than classes of logic-driven fuzzy neurons. The basic formulas
ordinary fuzzy control and the approach is practicable, simple governed the functioning of these elements are constructed
and effective.
with the aid of T norm and S norm (see Fig. 1, Fig. 2). Some
I. INTRODUCTION definitions of the double fuzzy neurons show their inherent
Fuzzy neural network combines the theories of fuzzy capability of reducing the input space.
logical and neural network, including learning, association, Definition 1. Let X =[XIX2...... Xn] be input variables and
identification, self-adaptation and fuzzy information process. W =[w1,W2 ....... wn ] be adjustable connections (weights),
The logic neurons have received much concern all the time as which of them are confined to the unit interval. Then,
the important components of neural networks. From the n
models designing to the algorithms studying, there are many y= T(w1Sx1) (1)
achievements. Glorennec [1] proposed a general artificial is the AND fuzzy neuron, which completes a T-S norm
neuron to realize Lukasiewicz logical operate. Yager [2] composition operators shown in Fig. 1.
employed a group of OWA fuzzy aggregation operators to Definition 2. Let X .[X ,. Xn] be input variables and
form OWA neuron. Pedrycz and Rocha [3] proposed
aggregation neurons and referential neurons by integrating W =[wI,W2 .......wn] be adjustable connections (weights),
fuzzy logic and neural network and discuss the relation about which are confined to the unit interval. Then,
the ultimate etwork structure and practical problem. Pedrycz n
[4-5] constructed a knowledge-based network by AND, OR S(wiTx) (2)
neurons to solve classified problem and pattern recognition. is the OR fuzzy neuron, which completes an S norm and T
Bailey [6] extended the single hidden layer to two hidden norm composition operators shown in Fig.2.
layers for improving complex modeling problems. Pedrycz AND and OR neurons both are the
and Reformat designed fuzzy neural network constructed by mapping [0, 1]n _> [0,1], the neuron expression is shown by
AND, OR neurons to modeling the house price in Boston [7]. Zadeh operators as (3) and (4).
We consider this multi-input-single-output (MISO) fuzzy A
logic-driven control system based on Pedrycz. Pedrycz [7] T -
transformed T norm and S norm into product and probability
operators, formed a continuous and smooth function to be T
optimized by GA and BP. But there is no exactly symbolic S,
expression for every node, because of the uncertain structure. t /, t
In this paper, the AND-OR FNN is firstly named as AND-OR W1/W2 \Wn W//2N
fuzzy neural network, the in-degree and out-degree of neuron x1/ I2 x 1 2 Xf
and the connectivity of layer are defined and Zadeh operators
are employed in order to infer the symbolic expression of Fig. 1 AND neuron. Fig. 2 OR neuron.
every layer and form a continuous and rough function. The

1 -4244-0529-7/06/$20.OO ©2006 IEEE

605
 y - A\WiVV Xi (3) XIg

y - V(Wi AXi) (4)

Owing to the special compositions of neurons, for binary *7
inputs and connections, the neurons' function is equal to the Zd
standard gates in digital circuit. For AND neuron, the closer to Xn
0 the connection w, is, the more important to the output the n

corresponding input x is. For OR neuron, the closer to 1 the
connection w. , the more important to the output the
t ~~~~~~~~~~~~~~~~~~~Fig.
corresponding input xi is. Thus the values of connections
become the criterion to eliminate the irrelevant input variables
to reduce input space.
B. Several notations about AND, OR neuron
Definition 3. Let wi be the connection value, xi be the ith
input variable. Then,
RD(xi) = wi E [0,1] (5)
is the relevant degree between the input xi and the neuron's
output. For AND neuron, if RD(x1) = 0, then xi is more
important feature to the output; if RD(x1) 1, then xi is more
irrelevant feature to the output, it can be cancelled. For OR
neuron, vice versa. Thus the RD V and RDM is the vector and
matrix made up of connections, respectively, according to the
dimensions of X, which has been become the threshold to
obstacle some irrelevant input variables and reduce the input
space. .
ef
Defilnition 4. In-degree
I iSi from directed graph in graph
theory; the neural network is one of directed graph. The in-
degree of the ith AND neuron d + (AND.
degree) isofteihADerndA
shown the number
ssonhubr
of his input variables; the in-degree of the jth OR neuron
d+ (ORj) is shown the number of his input variables.
Definition 5. The out-degree of the ith AND neuron
d- (ANDi) is the number of his output variable; the out-
degree of the jth OR neuron d (OR1 ) is shown the number of
his output variables.
C The architecture ofAND-OR FNN
This feed-forward AND-OR FNN consists of five layers
(Fig.3. MISO case), i.e. the input layer, fuzzification layer,
double hidden layers (one consists of AND neurons, the other
consists of OR neurons) and the defuzzification output layer,
and corresponding to the four parts (fuzzy generator, fuzzy
inference, rule base and fuzzy eliminator) of FLS (Fuzzy
Logic System), respectively. Here the fuzzy inference and the
fuzzy rule base are integrated into the double hidden layers.
The inference mechanism behaves as the inference function of
the hidden neurons. Thus the rule base can be auto-generated
by training AND-OR FNN\in virtue of input-output data.
input fuzzification AND OsR defzzification
3 Architecture of AND-OR FNN.
Both W and V are double connection matrixes, also
imply relevant degree matrix (RDM) like introduce above.
Vector H is the membership function of consequents. The
number of neurons in every layer is n , n x t, m, s and 1,
respectively (t is the number of fuzzy partitions).
Definition 6. The connectivity of layer is the maximum in-
degree of every neuron in this layer, including the double
hidden layers,
Con(AND) = max(d+(ANDi)) (6)
Con(OR) = max(d+(ORj)) (7)
where AND. is the ith AND neuron, d+(AND1) is the in-
degree of the ith neuron. OR is the ith OR neuron,
d+ (ORj) is the in-degree of the ith neuron.
Remark 1. Con(AND) < n ( n is the number of input
variables). Con(OR) < m (
ners)
m is the number of AND
neurons).
D The exact expression of every node in AND-OR FNN
According to definitions and the physical structure
background, the AND-OR FNN model is derived as Fig. 3.
The functions of the nodes in each layer are described as
follows.
Layer 1 (input layer): For every node i in this layer, the
input and the output are related by
= (8)
where Of denotes the output of the ith neuron in layer 1,
i = 1,2, ,n . Here the signal only transfers to the next layer
without processing.
Layer 2 (fuzzification layer): In this layer, each neuron
represents the membership function of linguistic variable.
Gaussian is adopted as membership function. The linguistic
value (small,, very large) A, are used. The function is
shown as.
show as2
-AJ(x me(9
(9)
where i=1,2, ..n, j=1,2, .t , mijand ufare the modal and
spread of the jth fuzzy partition from the ith input variable.
Layer 3 (AND hidden layer): This layer is composed of
AND neurons. Based on definitions above, the function can
be expressed as follows.

606
 d+ (ANDk ) d+ (ANDk) where xl,--, x, are the input variables, y is the output, A
0>k T
PI= (wO) A
PI
UA
(r
(Wk V Pij (1)
(Xi)) (10) and B are the fuzzy sets of input and output, respectively. wj
where k =1,2,-2 ,tt, j 1,2, --,t i =1,2, -,n r =(i-1)xt is local weights of the antecedent part; vl,-,vs is the class of
+j, d+(ANDk) is the in-degree of the kth AND neuron. tn is global weight for every rules. i =1,2,- -, n , j =1,2,- - ,t
the total of AND neurons in this layer. I = 1,2,, s , t is the total of antecedent fuzzy sets; s is the
Note 2. If p, is fixed and d+(ANDi) . 2, i must be different. total of consequent fuzzy sets. The same B1 is collected to
That means the input of the same AND neuron must be from form s complex rules as follows:
the different xi Rule 1: if xi is Aji,and and Xn is Ani,, then y is B,
Layer 4 (OR hidden layer): This layer is composed of or-- -or if xi is Alj, and-- -and xn is Anj, theny is Bl,,---
OR neurons. Based on above definitions, the function can be
expressed as follows. IRule S: if x1 is Akand --and x is Ank then y is Bs
d+(OR,) d+(ORi) or -or if xi isAn,, and .and Xn is Anl, theny is Bs
0> = S (vTO3)
P2='
v
P2='(/
I
(Vlk A O)(5 (1) (15)
where k =l,2,Xtn / =l,2,**,s d +(OR) is the in-degree of where i ', i ", j ', j", k', k",',I" E [1, t] .
the lt OR neron.sis thetotal f OR nuron.the firing On the view of general fuzzy inference, to a single rule,
( weight) of
strength (or i rule iS usually obtained
o ith
Layer 5 (defuzzification layer): There is only one node in . min or .s
this layer, which includes forward compute and backward by multiplication operator, to the complex rule, the
training. Center-of-gravity method is adopted for the former firing strength A 1is the union from all single weights, which
compute as follows. can be expressed as follows.

z= 05 = (12) Al= I(1UA,i,X ,Ani, (XI, SXn)1.. PAjIx .. xAn, (xI,- -Xn))
I

where 1,2,
-2 - , s , hI, is the membership function of Al j( (X-) UA(Xn))
A AP n, (16)
consequents. The latter only imports the training data for = V

directions are time-sharing. The function can be expressed as PP2=' ni

follows. where A, is the number of the same consequent B1, Ti is the
-(zd -mj)2 number of antecedent parts in a rule.
°5 =/1B (Zd) e
2
(13) On the view of fuzzy weighted Mamdani inference, local
weight and global weight are considered by a mode of
III. FUNCTIONAL EQUIVALENT BETWEEN THE < V, A > as follows:
FUZZY WEIGHTED MAMDANI INFERENCE AND THE
AND-OR FNN VI A i 'H
This section demonstrates the functional equivalence
A
Al VJPAni, (xn) )
yPI nWi (
between the AND-OR FNN and the fuzzy weighted Mamdani } Kl (WJJ V/IA (1x,--- (17)
inference system, though these two models are motivated VA, P l=1 V -(X) ))
A A X
from different origins (AND-OR FNN is from physiology and Wnjp Anj
fuzzy inference systems is from cognitive science), the In general, center-of-gravity method is adopted for
functional equivalent under minor restrictions is illustrated defuzzification as follows.
A FuzzyweightedMamdani inference s

The fuzzy weighted Mamdani inference system [9] Al PB, (Y)

utilizes local weight and global weight to avoid a serious y 1=1 (18)
shortcoming in that all propositions in the antecedent part are
assumed to have equal importance, and that a number of rules /=1
executed in an inference path leading to a specified goal or the B Functional equivalence and its implication
same rule employed in various inference paths leading to From (8) and (18), it is obvious that the functional
distinct final goals may have relative degrees of importance. equivalence between an AND-OR FNN and a fuzzy weighted
Assume the number of the fuzzy IF-THEN rules with Mamdani inference can be established if the following is true.
consequent y is B1 , then fuzzy rules can be represented as: 1) The number of AND neurons is equal to the number
Rule:ifx1 is Al.and,--and x is A.. of fuzzy IF-THEN rules.
and--,ad ~ i A1 ten (14) 2) The number of OR neurons is equal to the number of
and ,ad n i An ten i B1 fuzzy complex IF-THEN rules.

607
 3) The T-norm and S-norm used to compute each rule's
firing strength is min and max, respectively.
4) Both the AND-OR FNN and the fuzzy inference
system under consideration use the same center-of-gravity
method to derive their overall outputs.
Under these conditions, when RDM(wj) = 0 and
RDM(v ) =1, the AND-OR FNN is completely connected,
IJ
which is shown every part of antecedent is the same important
to the output,toWkte =n, / l= nxtt and
ouput,TA n
equal to th
nd eual o general
the genral
fuzzy inference. But it falls short of the practice. When
wi e [0,1] , vi e [0,1], every part of antecedent has different
important to the output and every rules has different important
to the final output.
IV. OPTIMIZATION FOR THE AND-OR FNN
In general, gradient-based optimization is often chosen
for ANN, Pedrycz [8] has proved that the output of AND
neuron and its derivative are heavily affected by the
increasing values of " n ". Meanwhile, in this paper the final
output of AND-OR FNN is not a smooth function. Obviously
gradient-based optimization is not preferred, A far more
comprehensive optimization policy is proposed, which is a
hybrid learning methodology comprising of three fundamental
techniques, namely genetic optimization, pruning algorithm
and ant colony system.
A Structure optimization
The key part is the structure design to solve the problem
using network, the AND-OR FNN could be over
parameterized, and in this case, they produce an over trained
net, which leads to worse generalization capacity. That is the
reason why, in order to eliminate unnecessary weights or
parameters. There are no certain methods to solve the
structure optimization. In this paper, GA (Genetic Algorithm)
and PA (Pruning Algorithm) are adopted to optimize the
structure for avoiding local optimization and reducing the
connections.
a. GA optimization
GA ( Genetic Algorithm) is a very effective method to
solving the structure optimization problem, the superiority is
robust, random, global and parallel process [9]. Structure
optimization is transferred to the species evolution by GA to
obtain the optimal solution.
1) Structure code mode
Parameters encoding is the first phase of GA
optimization. Parameters are often transformed to unsigned
binary. Here the structure parameters of two hidden layers are
consid Here on focus bye mat or as follows:
conu
d l
mtri
anfd]ll
RDM(AND) LWi,j imx(nxt) ( 9)
RDM(OR) =FV0r. 1 (20)
L"J lsxm
where i, j, m, n, t and s is the same like before. Lining up the
twvo matrixes to a binary string as initial population represents
the structure status, which is exactly satisfied to GA. In the
608
initialization, RDM(w,) 0 and RDM(vj) , the status of
AND-OR FNN is completely connected. The authoritative
opinion from experts can be put into initial population as
seeds to reach the best initialization.
2) Objectivefunction
Proceeding with the numeric data, we carry the genetic
~~~~~~~~~~~~~optimizationof the network. The parameters of the GA canbe be
chosen exerientally The fitness function
ch on . The
experimentally. function maximized
maximized by the
GA is expressed in the form of the sum of squared errors
between target values (data) and outputs of network. This
fitness has to be minimized (it stands in contrast with the
classic way of using fitness functions in GA whose values are
maximized; if necessary a simple transformation of taking the
reciprocal of the fitness, 1I/ fitness, brings in line with the
way of maximizing the fitness).
1
fitness (21)
( (zd ) )
b. Pruning algorithm
The pruning of the network is a process consisting of
removing unnecessary parameters and nodes during the
training process of the network without losing its
generalization capacity. The best architecture has been sought
using GA in conjunction with pruning algorithms. There are
three situations as follows:
1) For AND neuron, if the connection is equal or close
to zero, this means that the corresponding input impacts the
output of this neuron, the connection can be pruned away if its
value is close to one. For the OR neuron a reverse situation
occurs: the connection close to one implies that the specific
input is essential and affects the output of the neuron. The
connection can be pruned away if its value is close zero.
2) For RDM(AND) , the number of zero in every line is
shown as the in-degree of this AND neuron, the node can be
pruned away if the line is full of one. For RDM(OR) , the
number of one in every line is shown as the in-degree of this
OR neuron, the node can be pruned away if the line is full of
zero.
3) For RDM(AND) , if there are the same line in the
matrix, that means the node is redundant, can be pruned away.
For RDM(OR) , that means the nodes are conflict, need be
,,
pue wy
B. Ant colony optimization as parametric refinement
As well known, most neural networks are trained using
BP algorithm, whose main drawbacks are easily getting stuck
requires the output needing function longer training
in local minim and time. And it also
is smooth, v, A operators can't
satisfy at all. ACO (Ant colony optimization) algorithm [10]
was adopted to optimize the connection value of AND-OR
FNN here. ACO algorithm was proposed by Dorigo et al.
(1991). The main features are high-precision solution, fast
search speed, convergence to global optimum and greedy
heuristic search. The basic idea iS: assuming that there are m
parameters consist of the connections from GA and PA. First,
all the parameters are sequenced and set to N randomly values
in [0, 1], and for the parameter p (I < i < m), compose a set,
denoted as And then the number of ants is set to h, they
start from the random connection to search the optimal
solution, choosing next connection from every set
according to the amount of pheromones as follows:
J ))
Pr ob(r< (I i
=
N
(I )
Z Tg (Ipi)
g=1
where r1j (I'n )Piis shown the amount pheromones of the jth
connection pj (IP)i
When an ant finishes choosing the connection in all the
sets, it arrives at the food source, and returns its nest through
the paths just gone by it, and then adjust the amount of
pheromones corresponding to the chosen connection as follow
rules.
rj (Ip )(t+ m)= p-a (Ip )(t) irj(pi
+
hy(Ip) k =
Zk=1
where p(O < p < 1) is shown the evaporating of pheromone,
<Ti (IPi ) is the kth ant leave pheromones on the jth connection
in set It is described as follows:
ifpjJ (IPi(24)
) is selected
r1 (In.) = rQ/ek
LOi otherwise
otherwise
IZd
k
where Q is the constant to adjust factor. e = - z is shown
the network output error with a group of connection value,
which is selected by the kth ant. Obviously the less the error
is, the more the increment of the corresponding amount of
pheromone is.
Finally, all the ants find the best optimal solution for the
connections value.
V. SHIP CONTROL
According to the principle of manual steering, this paper
makes full use of the advantages of fuzzy logic and neural
network, and an AND-OR FNN controller is presented to
adapt the change of navigation environment and get the
information of ship maneuverability automatically. The
structure and parameters of AND-OR FNN controller are
learned by GA, PA and ACO. Fuzzy rules can be auto-
obtained from a group of sample data, which is generated by a
fuzzy control system. Fig.4 shows a block diagram of the
AND-OR FNN autopilot. Test results show that an effect
d(k)
A
NN k,d
- r(k) utopi(ot + 1)
(22)
Delay operator
Fig. 4 AND-OR FNN autopilot for ship course-keeping.
method has been provided for the improvement of ship
steering control.
Double input variables of the AND-OR FNN ship
controller are heading error A v = V(k + 1) - yi(k) and the yaw
rate y(k)= dil/ dt . The control action generated by the
autopilot is the command rudder angle 3(k) . The range of
values (universe of discourse) for a given autopilot inputs and
output are (-200,200) and (-2.5° / sec, 2.5°/ sec), respectively.
(Ip3)
It is usually required that the rudder should move
from35° port to 35° starboard within 30s.
A. The source ofthe experimental data set
In order to obtain integrated and exact training data, a
fuzzy control system for a tanker is simulated with nonlinear
Nomoto model as controlling plant. The input gains of a fuzzy
controller (K=0.16sec-1,
includes double inputs, theT=1O0sec).
The fuzzy controller
error in the ship heading A yi(k)
and the change in that error y(k) Every input variable is
fuzzied into 11 triangular membership functions; thus form
121 pieces of rules. The output of the fuzzy controller is the
rudder input 3. We assume the tanker as a continuous time
system controlled by the fuzzy controller, which is
implemented on a digital computer with a sampling interval of
T=0.1sec. From the view of manipulating index, it satisfied
the index of 150OOtdw tanker [11], a fuzzy closed-loop
controlling system is designed with fuzzy rule base from
experience. The positive direction angle is defined that the
rudder angle goes around toward right. Center-of-gravity is
adopted as defuzzification methods.
The reference input heading V'r is increased 45 degree
every 2000sec, rapidly. The digital simulation is completed to
closed-loop with the sampling periods h=1Osec, the sampling
state variable A y , y and controlled variable 3 , the ship
htading and desire si hea isshw insFig. 5tandzwe
obtinth
conventonal
chang
fuzzy oftrud.
control. The rstaseutng
age data sequence
609

20 ,RDiA '+i.and
Tiruw.(sec)
2.ft.wo
.............
lo 35X0
. ..
Fig. 5 Ship heading (solid) and desired ship heading.
B. The structure design and simulation results
In this section, an AND-OR FNN ship controller is
constructed successfully. Firstly, A V(k) and y(k) are double
input variables; 8(k) is the output variable, which are all
fuzzied into 3 Gaussian membership functions with 0.5
overlap degree, thus there are 9 pieces of rules. According to
analysis before, there are 9 AND neurons and 3 OR neurons
inin~ADO
AND-OR FNT. Intaiesrcueprmtr
FNN. Initialize ihal
structure parameters with all
connections to form the structure string that
is RDM(w ) = O, RDM(v,) = 1. The input-output data is from
{A v(k), y(k), 8(k), k =1,2,...} to train and test. The structure
and parameters are optimized by GA, PA and ACO, some
parameters are chosen experimentally, which are including
population size=100, crossover rate=0.7, Mutation rate=0.01
and selection process is tournament, and the parameters of
ACO are m 19 , pi E [0,1] , N 30 , p 0.65 , h 30 ,
= = = =
Q = 10.The ultimate better structure result is transformed into
the AND-OR FNN shown in Fig. 6. It is obviously that there
are 5 AND neurons and 3 OR neurons left, that is, 5 pieces of
fuzzy rules and 3 complex rules. The ultimate better
performance result is compared with fuzzy control system.
Test data set is form the different part with training data in the
same sequence introduce above. Fig. 7 shows the error
comparison between test results and objective data. The
simulation result illustrated the effectiveness of proposed
method.
AV(k)..0C 04 Wp >
/JfQ\\\V
~
Y(ki ~ ~ ~ -z
P 5 >
Fig.6 the AND-OR FNN ship controller structure.
_~v77-7~7
Fig. 7 The performance index compared.
VI. CONCLUSIONS
In this paper, we have proposed a novel AND-OR FNN
a piecewise optimization approach, the symbol
= expressions of every node in the AND-OR FNN are educed in
detail, the input space is reduced by special inner structure of
AND and OR. The equivalence is proved to the fuzzy
weighted Mamdani inference. The fuzzy rule base auto-
extracted is equal to optimize the architecture of AND-OR
FNN. This novel approach has been validated using AND-OR
FNN ship controller design. The simulation results illustrate
the approach is practicable, simple and effective and the
performance is much better than ordinary fuzzy controller.
ACKNOWLEDGEMENTS
Edcto
This paper is supported
Cmites
by the Doctoral Foundation
203505 Mnsry
off
Education Committees (2003151005) and Ministry of
an
Communication of P. R. China (200332922505).
REFERENCES
[1] G. Pierre Yves, "Neuro-Fuzzy Logic," In: Proc. IEEE-
FUZZ., New Orleans, pp. 512-518. 1996.
[2] R. Yager, "OWA neurons: A new class of fuzzy neurons,"
In: Proc. IEEE-FUZZ., San Diego, pp. 2316-2340. 1992.
[3] W. Pedrcy and A. F. Rocha.., "Fuzzy-set based models of
neurons and knowledge-based networks," IEEE. Trans.
on Fuzzy System, vol. 1, no. 4, pp. 254-266, 1993.
[4] K. Hirota and W. Pedrycz, "Knowledge-based networks
in classification problems," Fuzzy Sets and Systems, vol.
159, no.3, pp. 271-279, 1993.
[5] W. Pedrycz and G. Succi, "Genetic granular classifiers in
modeling software quality," The Journal of Systems and
Software, vol. 76, pp.277-285, 2005.
[6] S. A. Bailey and Y. H. Chen, "A two layer network using
the OR/AND neuron". In: proc. IEEE-FUZZ., Anchorage,
pp. 1566-1571, 1998.
[7] W. Pedrycz and M. Reformat, "Genetically optimized
logic models," Fuzzy Sets and Systems, vol. 150, no. 2,
pp. 351-371, 2005.
[8] D. S. Yeung and E. C. C. Tsang, "Weighted fuzzy
Z/ - -production rules," Fuzzy sets and systems, vol. 88, pp.
299-313, 1997.
[9] S. L. Zeng, L. J. Song and W. H. Xu, "Neural network
structure optimization based on genetic algorithm,"
Journal of Jishou University, vol. 26, no. 3, pp. 118-120,
2005.
[10] H. B. Duan, "Ant colony algorithm: theory and
applications," science press, 2005.
[11] G. X. Yang, "Study on ship motion hybrid intelligent
control and its interacting simulation based on virtual
reality," PhD Thesis. Dalian: Dalian Maritime University,
610