Scribd
Upload
54 views12 pages

SC 2 PDF

This document describes the architecture and computation process of an Adaptive Neuro-Fuzzy Inference System (ANFIS). ANFIS takes a Sugeno-type fuzzy inference system and constructs an equivalent ANFIS architecture. The architecture consists of 5 layers - layers 1 and 3 contain adaptive nodes, layers 2 and 4 contain fixed nodes, and layer 5 computes the overall output.

Uploaded by

MONTY
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
54 views12 pages

SC 2 PDF

This document describes the architecture and computation process of an Adaptive Neuro-Fuzzy Inference System (ANFIS). ANFIS takes a Sugeno-type fuzzy inference system and constructs an equivalent ANFIS architecture. The architecture consists of 5 layers - layers 1 and 3 contain adaptive nodes, layers 2 and 4 contain fixed nodes, and layer 5 computes the overall output.

Uploaded by

MONTY
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 12

Dt2012.

22

Problem
ackprcpagatton alqarcthm
a1o a2
T-Ql
o.5 Outpa
0.9
Input Hidaen
Uprlare he eight pto one tkation)
A Step-l
Toput layet Camputaho

-0,9
Siep2

o2 Jgy2
0505
The total mput vector

01 0.2JL-2x 04X0.1 t o.2xfa ?)


Q18 7 iput to hidcen
_0.02J ayes
utput of hiddan lau t
14eAO18) Assone =
Ove
-14E O64IS
LO.505
He
SHo-3
Quhut la yer Computaton.
=w'Q aput of ouput layor

0205 515
20,505 Jayal
Q2x9511S -05)xasas]
-0134

-0.134 Output of adput layor

046
*Cmpute the Erøror

= O066
pdate the cciabt
new

Aw
ncwolaAV f thee b muhiple

As H ba zie neuc

Let 2a6
O.515

d ) 1-0)
=
1-0,464) Q461Cl-a164)
-0AO5

d-0.09as
Aw= .
0.6o.5t45 xt90905
0.505-x Vecar mltilicat an

-00296)
-0.029 old
New +Aw

-a0216
-o02
4chaie V

0.6
SE
SV
=
<t7 T

ta

e j the
swbonpt
o new t hicben

Y Cfixed
e7.
K

K02
0005 x 02 x 0511|- a5115) x
=0,218
W-05
K
)=-0.oq05 X(-05) X0505 (I 0505)
0,0
0.48
LO, ol

O.Ga1 o,248 00oll)


a
Dt-83122

Kadial Basd funchon Netwartks (RBFN)


THs a cla» of eed otword netwsrtka .

The hidden neuron


achvahons m RBFN ate
cmpuktd us
the Exponential dstance meauce bettueen the hput and
o prototype vecder ha Chaacerise, the sjmal foctn
at a hidden newron,
Thes clane of netuork ane many wed tr luhepolehon,

hechiteckure d RBAN

N
na of
feafwes
Oncn a trahhy Aet T {XdJ
N-noo trahmy samplcs
E R, d eR
Oljechue
fmd a
funchon or a
mop f(:) duch that

K=, .N -daed o
far fnda f uhich best naps
it mput vector b i
deswred otput, the RBFN
CoB?deco apumex a cla» of
Nbasfunctons
hon-laearr hnchon and
(1X-X11) tr the tunchan daa
takes the Euchdecn
betueen ctonce. la
an
mput vector X ad teahmq cate
Pomt .
k-L2. ...N
Usthe Basis fuchon equahon
an be ettten

k=L3N

D i s one output vectá, then Cayde Coy


T
D-4. Irranpoe
TWis ueiar vecor
W
Thebass tuction madretix ib

Ustnq W, D and , eg Can be retrHen as-

W=D Dis a Symnchdc


from eg",W can be camputed a Dimcaoion- ndra.
Nxy
o-o a Symctrt*
w- D
Ttobrercved that for a laqe claw of
is
furchons, o
Non-squlart przonided the traning data polits are
cisthct and hence o Can be ivocftble. Thee class af
tunctto ate-
(D Giaussian tuncthon
w- expl-6-c C-Ccoter
teod
Multiquadrat tunchon

nvotse functon
()=

Choosm Cin qpprcoprtkale emures its hvectibil

The Conmonly wed functn fr is Oawskn uncfon


This funchton s ako alled Racal Basts Fuchon arc
as

herce the nane ofthe nttark is kacial 6ass funcho


Netozk[RBAD.
ANFIS fdaplive Newo fuzz Intorencepm)
ANF1S is a clas of Neural
Nettuonk tich rk, a a
fuzzs
nkcrere ydem
To ariale tuo mput Sugeno fuzzy madel
To consituct an TCt take
ANFIS, otaka a tuo vartialle tuo
mput
Suqeno-fuzz Madel whore the rules are mhe
Rule : 1f
forrn-
x ib
A, and "thenB,
Guhere
f-Rxt9jt,
f, o are contants
Fule x, ard b& then
Cuhore
Jt
fa gare constants
To get he hforerce forr an put and y, fireat fhe
npus caill be apphed fo te, rpeche tezy ses of ech
ule, than the Concluion of a Kule tuill be
tuo sdeps,
Campaecd

outnt
Then the fmal Hput foc an mput is
Ri w
X

Xx
Arrchitecture

Computan of layer1
The necwca at layord ol
ANFls are AdapBve
newony
whore the itb node Compudes it's
output as-

-3,1
Compuctalion of layer2
The newon of layer-2 are
Hixed nodes ad
conpes hetre
Outpat au

Tor i2
Coputatton of layor-3
The newons at
lauor-3 are tixed neutns and the
oupet of
ecch neuon at layor 3 B dened a2
or f-l,
tu
Computaton of layer
The neurons arre Adopfive newtos
ond the
oflayor
ith newron of this
layer is

Canpufaton of layer 5
The sale node m thrs layerr a tixed nade, thich comoue
he overall output as fhe Summafon of all hce mput
demed a- 1

You might also like