Hg suy diln md tren ca sd mang thich nghi ANFIS co chiic nang tuong tu nhu hg ANFIS cdn sii dung cac luat hoc lai d l hoc cac ham thuoc, xac dinh cac tham so va sinh ra tap c4c chiing t
Trang 1I T N G D U N G H £ S U Y D I E N MOf (ANFIS) GIAI QUYET BAI TOAN PHAN NGANH CHO SINH VIEN KHOA CONG NGHE THONG TIN TRU^dNG DAI HQC CONG NGHE THONG TIN &TRUYEN THONG - DHTN
Nguyen Thj Tuygn*, Dao Thj Thu
Trudng Dgi hoc Cong nghe thong tin & Truyin thong • DH Thai Nguyen
T O M T A T
ANFIS (Adaptive Network based fuzzy inference system) la mot mo hinh he suy dien md dua tren
vdi muc dich 1^ tang nhOng uu diim va ban chS t6i da nhung nhupc dilm cua he suy dien md v^ m^ng noron Hg suy diln md tren ca sd mang thich nghi (ANFIS) co chiic nang tuong tu nhu hg ANFIS cdn sii dung cac luat hoc lai d l hoc cac ham thuoc, xac dinh cac tham so va sinh ra tap c4c
chiing toi da sir dung mo hinh nay trong bai toin phan nganh cho sinh vign Khoa Cong nghe thSng tin- Trudng DH C6ng nghg thSng tin &Truygn thong vdi mong mudn giiip cho cac em c6 duac ca
sd d i tu tin budc vao ng^nh minh da lua chon
T i ' kh6a: mgng noron, hi md, he suy diin md, ANFIS phdn ldp
DAT VAN DE
Ly thuyet md va mang noron nhan tao da phat
trign rat nhanh va da dang Chinh nhung cdng
cu nay da cung cap nhiJng cdng ngh? mdi cho
cac nganh cdng nghiep lam ra nhieu san pham
thdng minh, dap ung dupe day dii nhu cau ciia
thi trudng can cd nhiing bd digu khign linh
hoat hon, nhiing thigt bj "biSt suy nghi", lam
viec vdi nhirng bai toan kho, phai xii iy nhigu
thong tin md, chua diy dii va thieu chinh xac
Ly thuygt md va mang noron nhan tao dgu cd
nhung uu didm va nhugc didm rieng Chinh
vi vay, mot each tiep can chac chan se dem lai
nhilu thanh cdng va la hudng md cho nhigu
nha nghien ciiu, dd la tich hgp giiia he md va
mang noron l^i vdi nhau ggi la md hinh h?
suy diln md' dua b-gn mang thich nghi
(ANFIS) Viec tich hgp nay se lam tang thgm
nhiing uu dilm ciia ca hai he va khac phuc
dugc dang ke nhung nhugc dilm ciia ly
thuygt md va mang noron nhan tao
Do he ANFIS dugc tao thanh tir he suy diln
md va mang noron nhan tao, do vay vdi kha
nang hgc cua m^ng noron nhan tao cho phep
dieu chinh viec chuyen doi cac luat md if
-then va cac bam thugc kit hgp vdi cac y kiln
* Tel- 0988 808459, Email nttuyen@ictu edu v
chuyen gia se hiia hen dem l^i each giai quygt dem lai ket qua cao vdi cac bai nhu phan ldp,
du doan De cd thi md ta dugc kha nang ling dung cua he ANFIS, trong bai bao nay, chiing
ta di tim hieu mdt irng dung cua he trong bai toan phan nganh cho sinh vign Day la mdt ling dung dugc dl cap rit nhilu trong cac cupc hdi thao Idn, ddng thdi nd cung dem lai rit nhieu hieu qua va thanh cdng trong nhieu ITnh vuc dac biet la Y hpc va ITnh vuc dieu khign,
Bai bao cd cau true nhu sau: Sau phan md diu, bai bao trinh bay cau triic, each thiic hoat ddng cua he ANFIS Phan kg tiep trinh bay kgt qua thii nghiem tren bai toan phan nganh cho sinh vien thudc Khoa Cdng nghe thdng tin- trudng DH Cdng nghe thdng tin
& Truygn thdng Cudi ciing la kgt luan va thao luan
KIEN TRUC HE SUY DIEN M 6 D U A TREN MANG THICH NGHI (ANFIS)
De hieu rd ve mang ANFIS, trong phan nay chiing ta se dua ra cau true cua mang thich nghi tuong duong vdi he suy dign md trong
do chiing ta se di sau vao md ta each phan chia cac tap tham sd de ap dung luat hpc lai
va each thuc hien cac luat md if then[S\
Trang 2Uyer 1 Layer 2 Layer 3 Layer 4 Layer 5
i i i
X ? h - '
De don gian, ta gia su rang he suy dign md cd
hai dau vao x va y va mdt dau ra Gia sii rang
cd hai luat co sd md ;/ - then ciia Takagi va
Sugeno nhu sau:
Rule 1 : \i X h Aj and y is Bj, then // =
pix+qiy-^r,
Rule 2 • If x is -^j and y is B2 then /^ =
P2x+q2y+r2
Cac niit trong ciing mdt ldp cd chuc nang nhu
nhau nhu md ta dudi day:
Ldp 1: Mdi mit i trong ldp nay cd hinh vudng
vdi cac ham thanh vien
0'(x) = fiA,(x) vdi / = 1
2-l ' o a c o / ( ' x ; - / i f f , _ / x ;
vdi , = 3 4
6 day, X (hoac y) la diu vao nut i va A^
(hoac B,_2) la cac bien ngdn ngii kgt hgp vdi
nut ham nay Hay ndi mot each don gian O'
la ham thudc ciia A, va nd chi ro dp thudc
ciia X vao A^
Thudng ta chpn /iA/x)\k ham thudc hinh
chudng vdi gia tn Idn nhit la 1 va nho nhit la
0, nhu vay ham chudng tdng quat se dupe md
ta dudi dang toan hgc nhu sau:
M A , X ) ^ ^ •
Trong dd : a„ b„ c, la tap cac tham so Khi gia
tri ciia cac tham sd nay thay ddi thi ham sg
bign ddi theo Cac tham sd trong ldp nay la
tham sd gia thigt
Ldp 2: Mdi mit trong Idp nay cd hinh tron va
cd nhan la FI N o chinh la tich ciia cac tin hieu den va moi gia trj dau ra ciia mit bieu dien cudng do ciia mgt luat (toan tir T - norm thuc hien phep AND tdng quat cd thg dugc sii dung d day)
Of(x) = w, = ^A,(x).piB,(y),i^l,2
Ldp 3: Mdi mit trong Idp nay cd hinh tron co nhan la N Niit thu i dugc tinh la ti le cua cudng do luat ciia mit vdi tdng tat ca cac cudng dp luat:
Dg thuan tien, gia trj dau ra cua mit nay se dugc gpi la gia tri dau vao lufit dupe chuan
hoa {normalized firing strengths)
Ldp 4: Mdi ldp i trong ldp nay la mgt nut hinh vudng cd chiic nang
Oi(x) = -Wj^ ~ W,(p^ +^,y-^>':)
Trong do, w^ la dau ra cua Idp thii ba
[p,' 9i' ^ } 'a tap cac tham sd , cac tham s6
trong ldp nay se dugc xem nhu la tham s6 ket luan
Ldp 5: La mgt Idp cd mot nut hinh trdn, hong Idp nay cd ky hieu la Z la diu ra bing tong ciia tat ca cac ky hieu din
overall
oulpul = o;(x) = -^,w,f, :
Z.",
Chiing ta CO thj cau true mot he ANFIS tuong duong voi mot he suy dien mo voi cac luat
ma TSK Kiin true cua he ANFIS CO thS cap
nhat cac tham so theo thuat toan lan truyk
Trang 3ngugc Neu khdng ta cd thi sii dung thuat
toan Ipc Kalman dung dl tim cac tham sd kit
lu|in trong he ANFIS, dieu nay cd dugc bing
each la ta sip xip tit ca cac tham so kit luan
vao mdt vecto (r/, pi, q,, r2, pi qi) , sau do sii
dung thuat toan Ipc Kalman de giai quygt tit
ca cac rang bugc theo phuong trinh tuyen tinh sau day:
—fll
w,
—1')
Wl
—('><!)
Wl ^
—' ^t!)
Wi^
—"> ,,(•) w, y
w,y
—to
W2
—1!) W2
—111(1)
Wl ^
—"><2)
Wl ^
_Wi
Trong do [x"' f>, I
WiV'
-IHydl' w^y'"
'•i
PI
II
1-2
Pi
1 2
_
'b<•>^ b">
b"\
la dau ra d Idp
a cap du lieu huan luyen thii k, k = 1,2, , p wi
thii ba, ket hgp vdi dau vac (J:*, y'''')
Gia sii rang chiing ta su dung hit tit ca cac kha nang ciia cac tham sd la sd lugng dau vao va sd lugng cac ham thugc khi dd cd the dinh nghTa sd lugng cac luat la:
Rule
-nw
Va neu tham sd gia thiet (premisparan) la sd lugng tit ca cac tham sd can cho ham thupc, khi do
sd lugng tat ca cac tham sd dugc dinh nghTa la:
para,, = premispara,^'Y^ln„Mf, +RiileJIn„ -\-l)
i^t
THUAT TOAN HOC LAI VA XAC DINH THAM SO THICH NGHI
Thuat toan hoc lai
D I CO thi dem lai hieu qua cao nhit trong qua trinh tinh toan, he ANFIS su dung thuat toan hpc lai Tu kien true mang ANFIS da cho trong hinh (2.1), chung ta thay rang neu cho gia tri cac tham sd gia thiet, dau ra cd thg dugc dign ta bdi mgt ham kgt hgp tuyen tinh cac tham sd kgt luan Noi mot each cu the, dau ra f trong hinh (2.1) cd thg dugc cu the hoa nhu sau[5]:
f=wJi+wJ,=(w,)r,+(WiX)p,+(w,y)q^^(w,)r,-v(w,x)p,+(w2y)q^
f la tuygn tinh trong cac tham sd ket qua (pi, qi^n, p2, q2, r?) Kgt qua chiing tacd:
S = tap ciia tdng cac tham sd
S[ = tap ciia cac tham sd gia thigt
S2 = t§p ciia cac tham sd ket luan
Cu the hon, trong pha di tdi ciia thuat toan hpc lai, ham tin hieu di qua bdn Idp (di den ldp thii 4) va tham sd kit luan dugc xac djnh bdi phuong phap udc lugng binh phuang tdi thieu (LSE) Cdn trong pha di lui, ti le loi dugc lan truyen ngugc va cac tham sd gia thiet dugc cap nhat bdi phuong phap giam gradient, cac boat ddng trong mdi pha boat ddng dupe tong quat hoa qua bang sau:
Bang 1; Hai pha trong thd tuc hoc lai cho hi ANFIS
Tham s6 gia thiet
Tin higu
Co dinh LSE Niitra
Giam gradient
Co djnh
Ti le loi
Trang 4Theo nhu bang tren, chung ta nhan thiy, cac
cdng thirc cap nhat cac tham sd gia thiet va
kit luan dugc tach ra trong luat hgc lai Do
do, viec tang tdc dp hpc la cd thg bang each
sii dung mot each khac cua phuang phap
giam gradient trong phin gia thilt nhu giam
gradient lan truyin nhanh, tdi uu hoa phi
tuygn
Tren thuc tl, vdi mdi phuang phap dgu cd
nhiing uu va nhugc digm rigng cua nd,
Phuong phap udc lupng binh phuang tdi
thilu va phuang phap giam gradient cung
vay Do do, ngudi ta d3 dua ra dupe mdt sd
each dl cap nhat tham sd (tuy timg trudng
hgp cu dg ap dung vi mdi mdt phuong phap
deu cd mdt dp phiic tap tinh toan nhat dinh
nao dd)
Chi giam gradient: Tat ca cac tham sd dugc
cap nhat bdi phuang phap giam gradient
Giam gradient va chi mdt lan diing LSE,
trong do LSE dugc su dung chi mot lan tai
thdi diem bat dau dat cac gia tri khdi tao ciia
cac tham sd kgt qua va sau dd su dung giam
gradient de cap nhat tat ca cac tham so
Giam gradient va LSE: Day la luat hpc lai
Chi LSE: He ANFIS la tuyen tinh va cac
tham sd gia thigt va sir md rgng thuat toan Igc
Kalman dupe dimg de cap nhat tat ca cac
tham so
Xac dinh tham so hpc thich nghi
Ddi vdi mdi niit trong mang thich nghi sau
khi da co cac ham thanh vien thi nd se khdi
tao cac gia trj cua cac tham sd gia thigt qua
ham thanh vien trffng mien gia tri cua cac
bign vao'''l Vdi mdi gia tri cho trudc x cua
mdt dau vao trong mien tinh toan, chiing ta
ludn tim duoc mdt bien ngdn ngu A sao cho
MA(^) - ^ vdi la ngudng cho trudc
Theo each nay, he suy dien md cd the tao ra
su bign ddi vira dii su chdng lap giiia cac biin
ngdn ngir Tham sd k anh hudng din din tdc
dp hdi tu, nlu k nhd thi cd the xip xi sat vdi
dudng gradient nhung hdi tu cham va tdn thdi
gian Ngugc lai, neu k Idn thi su hdi tu se
nhanh nhung ket qua se lan can xung quanh diem tdi uu, bing quan sat chiing ta cap nhat
k theo hai heuristic sau:
NIU dp do loi giam I ign tiep 4 budc thi tang k 10% NIU dp do Idi giam sau hai budc lign tilp thi giam k 10%
DSy la chiln luge c^p nhat ddng, do vay gia trj khdi tao k se khdng ddng vai trd quylt dinh nlu k khdng qua Idn
UNG DUNG HE ANFIS GIAI QUYET
B A I T O A N PHAN NGANH CUA SINH VIEN TRONG KHOA CONG NGHE THONG TIN (CNTT)
Thuc trang van de dang ki nganh hoc cua sinh vien
Trgn thuc tl, tat Khoa Cdng nghe thdng tin
-Trudng DH CNTT&TT, hang nam phai giai quylt nhilu dan xin chuyin nganh cua sinh vien Chinh viec xin chuyin nganh nay din dgn viec quan Iy Idp va quan ly dao tao ciia Khoa ndi rieng va ciia Trudng ndi chung tr6 ngn kho khan va pbiic tap Ly do ciia viec chuygn nganh nay phan Idn la khi cac em dang ki thi hoac xet tuygn (hoac dang ki
nguyen vong 2, nguyen vgng 3) vao mot
nganh trong trudng cac em diu khdng tim hieu kT thdng tin cua nganh dd de cd thi biet dugc cd phii hgp vdi kha nang, sd thich ciia minh hay khdng Cd rit nhilu Ii do cac em chpn lua vd thiic nhu: cd em thiy tgn hay hay,
cd em lai thay nganh nay lay diem thap, dg vao ,De giup cac em cd dugc su tu vin khi chgn vao nganh hpc ciia Khoa CNTT phii hgp vdi nang lire, sd thich cung nhu cac dilu kien vdn cd vg hpc lire, dam me, giiip cac em co hiing thu trong hgc tap sau khi triing tuyen va
cd viec lam thich hgp sau khi ra trudng, dong thdi giilp cac phdng ban chiic nang quan ly t6t dugc sinh vien chung tdi da dua ra each giai quyet bai toan phan nganh sinh vign trong Khoa Cdng nghe thdng tin sii dung he suy dien md tren ca sd mang thich nghi Cac biro'c ling dung he suy dien md tren cff
so mang thich nghi de giai quyet bai toan
Trang 5phSn nganh cho sinh vien Khoa CNTT
-Truwng DH CNTTTT - DHTN
De giai quyet dupe bai toan thuc tl nay chiing
tdi sir dung cac budc nhu sau:
Budc 1: L§p phigu khao sat va khao sat tinh
hinh thirc tr^ng trudc khi dang ki nganh hgc
ciia cac em sinh vign Trong tdng sd 550
phigu phat ra cho cac em sinh vien KIO va
Kl 1 trong khoa Cdng nghe thdng tin,
chimg tdi da thu vl dupe 526 philu khao sat
dap ling dugc yeu cau
Budc 2: Dua tren phigu khao sat da thu thap
dupe, chiing tdi tien hanh phan loai, riit trich
thdng tin, xir ly thdng tin dg dua ra duoc tap
dii lieu dau vao can xti Iy Sau khi thuc hien
viec phan loai phigu khao sat, sd phieu diing
dg kigm tra la 380 phigu
Budc 3: Sii dung he ANFIS de xu ly, huan
luyen dii lieu, dua ra kit qua
Budc 4: Tir kgt qua dua ra dugc nhan xet ket luan
KET QUA Chung tdi tien hanh chay chuong trinh bang ANFIS trgn ngdn ngff Matlab phien ban R2012a vdi tap dir lieu thu dugc tir 380 phieu dilu tra ciia sinh vign bao gdm cac thdng tin thdng qua nam vin dl chinh nhu sau: kit qua hpc lire mdn toan va ly cac nam THPT, thanh thao mdt ngdn ngu, kT nang lap trinh, van dd dam me tim higu, mong mudn dugc lam cdng viec Dua vao thdng tin ciia cac van de tren, sinh vien cd thg dugc dua vao 1 trong 4 nganh phii hgp vdi kha nang va sd thich cua cac em Ddi vdi tap dii lieu nay thi d&u vao ciia he thdng gdm cd 5 mit Dg ket qua dugc chinh xac, chiing tdi sii dung 200 bg dir lieu mau chia diu cho 4 nganh Sau 300 lan lap, kha nang chju Idi la 0, sd ham thanh vign lira chgn cho cac dau vao cd dugc nhd phuang phap phan cum Tiep theo chiing tdi sir dung 380
bg dif lieu da thu dugc de kigm tra viec dang
ki nganh ciia sinh vign
m 200 350 m 350
Hinh I: Trudc khi phdn ngdnh sinh vien Hinh 2: Kit qud phdn ngdnh ciia sinh vien
Nhan xet:
Tir hinh 2 kit qua cho chiing ta thiy rang, trong tdng sd 380 em sinh vien dugc dung de phan vao
4 nganh: Khoa hpc may tinh, He thdng thdng tin, Ky nghe phan mgm, mang&truyln thdng Trong do kit qua cd khoang 28 em chiem khoang 7,4% tdng sd sinh vign dugc dilu tra chua thuc sir dugc vao diing chuygn nganh ma cac em ygu thich hoac phii hgp vdi kha nang ciia chinh ban than cac em, Dilu dac biet la trong 28 sinh vien nay, cd tdi 17 sinh vign d rit xa so vdi muc tieu, dieu dd the hien ring viec lira chpn nganh ciia 17 em la hoan toan khdng phii hgp Cd thg do
Trang 6nhiSu ly do nhu tai thoi dilm dang ki mot s6 sinh vien chua co hieu biet ve nganh nghe ma minb lua chpn, hoac do phu thu$c vao dilm tning tuyln khi dang ki nguyen vgng 2, nguyen vong 3 ciia timg nganh la khac nhau Chinh viec tea chpn nganh khong chinh xac nay lam cho cac em khong hung thu, say me, chuyen tam vao hpc tap, din din kit qua hpc tap khdng cao, lam anh huong true tilp den ban than cae em
KET LUAN
Tren CO so nhiJng kit qua da dat dupe chiing toi CO thi kit luan rang vi?c su dung h? ANFIS vao bai toan phan lop, phan nganh cho sinh vien dem lai nhimg hi$u qua tot dep Ket qua ciia viec phan nganh giup eho eac em hpc sinh eo thi tham khao dl dua ra quylt dinh dung dan nhjt truoc khi dang ki vao nganh hpc tai khoa Cong ngh? thong tin - Tructng dai hpc Cong nghe thong tin & Truyin thong Tir viec chpn dupe nganh yeu thieh va phii hpp voi kha nang, cac em se yen tam hpc tap va nghien ciiru dl eo duac ket qua cao nhat
TAI LIEU T H A I V I K H A O
1 Bili C6ng Cuimg, Nguyin Doan Phudc, He mir, mang noron va frng dung, nha xuat ban khoa hpc va ky thuat, mn?i, 2001
2 Adriano Cruz, ANFIS; Adaptive Neuro-Fuzzy Inference Systems, 2003
3 Kazuo Tanaka, Hua O.Wang, Fuzzy Confrol Systems Design and Analysis, A Wiley-Interscience Publication, John Wiley & Son, Inc., 2001
4 Piero P Bonissone, Adaptive Neural Fuzzy Inference Systems (ANFIS):Analysis and Applications,
2002
5 Manish Kakar, H6kan NystrSom, Lasse Rye Aarup, Dag Rune Olsen, Respiratory motion prediction by using the adaptiveneuro fijzzy inference system (ANFIS), 2005
SUMMARY
APPLICATION OF A M I S TO SOLVE THE STUDY SPECIALIZATION
CATEGORIZATION PROBLEIM FOR THE FACULTY OF INFORMATION
TECHNOLOGY, COLLEGE OF INFORMATION AND
TELECOMMUNICATION TECHNOLOGY - TNU
Nguyen Thi Tuyen", Dao Thi Thu
College of Information and Communication Technology - TNU
ANFIS (Adaptive Network based fiizzy inference system) is a fiizzy logic model based on adaptive network This model is an integration of a fiizzy system and adaptive neuron network that
network Adaptive Network based fiizzy inference system (ANFIS) has the same fiinctions as a fiizzy system but overcome the disadvantages of such systems Moreover, ANFIS employs hybrid-leaming rules to learn memberehip fiinctions, to determine parameters and to generate if-then fiizzy rules to approximate the desired data Using the advantages of ANFIS, the authors of this paper applied this model to die study specialization categorization for smdents studying at the
to help shidents be confident in choosing the study specialization for diem
Keywords: neural network, fuzzy, fuzzy inference system, ANFIS, classification
Ngay nhdn bai:IWI0/20l4: Ngay phan bien:04/l 1/2014: Ngdy duyet ddng: 05/3/2015
Phdn bien thoa hoc: TS Vu Vinh Quang - Tnr&ng Dgi hoc Cong nghe Thong tin 4 Truyin thong - DHTN
Tel 0988808459 Email: ntluyen@!ctu edu.vr