Mô hình hỗn hợp và ứng dụng trong ước lượng các quan hệ phi tuyến

MO HINFl HON HOP VA UNG DUNG TRONG UOC LUONG CAC

QUAN HE PHI TUYEN

Phijm Thiivng I Idn, Trdn Ilodi Linh"", Nguyin Qudn .\'hn""

(a) Tririrng Dgi heic Bdch Khoa lid Ndi

(h) Trudng Dgi hge Ky thudt Then Nguyen

*

Tdm tdt:

Bed beio se trinh hdy ve nu}l md hinh me'ri etc live lugng ce'ic eptein h^ i^Iii tuyen giua ce'ic eteii

lire/iig ddu veto vd dau ra ciia mgt etdi tugng. Vi^^c sir dung phdi he/p cdc thudt toe'in udc lirgng

tuyen tinh vd ce'ic thudt toe'in ue'xc lugng phi tuyen etd giup ndng cao dugc chdt lugng ciia mo

hinh irirc lugng, gidm etuerc .sai sd dau ra. Md Innh cung sir elung ceic hi .so ciia md hinh tuyen

tinh ete.xdc etinh mire dg dnh hudng ciui cdc the'mg sd etdu vdo tai ket epid, tu etd cho phep lira

chgn dire/c ceic the'mg .so leit vd gidm mire dg phirc tgp cua md hinh tong thi. Cdc gidi pheip ly

thuyet .se dugc thir nghiem vai .sd lieu cua bdi le'iein eiu beio dinh ciia hiiu dd phu ldi ngeiy chu

khu vuc Thdi Nguyen.

Abstract:

The peiper will present abenit new model for estimating the nonlinear tran.sfer functions

between two set e)f signals. The simultaneous application e)f linear and nonlinear estimators

helps to improve the accureicy e>f the meielel. The hybrid model will find the he.st coefficients

for linear estimator and also use these coefficients in nemlinear estimator to reduce the

complexity e)f the moelel. The proposed method will be tested with the predjleni e)f P,„as

prediction for Thai Nguyen pren'ince.

I. DAT VAN DE

Bai todn dy bdo va udc lugng ndi chung la mdt trudng hgp dac biet cua bai todn xay dyng

md hinh dnh xa giira ddu vao va ddu ra. Da cd rdt nhieu md hinh udc lugng dugc de xuat » » »

[3,4], trong do su dung cdc phuang trinh tuyerl ti'nh hodc phi tuyen. Cac md hinh tuyen tfnh co

uu diem la don gidn, de dang tfnh toan va udc lugn§, ddng thdi ban chi dugc nhCmg anh

hudng tieu cyc cua cdc nhieu tdn sd cao. Tuy nhien ddi vdi nhirng ddi tugng phiirc lap co do

phi tuydn cao tbi sai sd cua md hinh tuyen tfnh se Idn. Khi dd ta cin sir dung cac md hinh phi

tuyen. Khi dugc lya chgn phii hgp thi sai sd dat dugc se thap. Nhung khi sir dung cac md hinh

phi tuydn, ta se gap mdt sd cdc khd khdn ca bdn sau:

Viec lya chgn dang ham truyen dat: do cd rit nhilu cac dang ham truyin dat phi tuyIn

khac nhau nen viec lya chgn dang ham phii hgp se phirc tap horn nhieu so vdi trudng hgp

hdm tuyen tfnh.

Vide udc lugng vd xdc djnh cdc thdng sd cua md hinh phi luyen: Qua trinh xac dinh cac

thdng so ciia md hinh phi tuyen la mgt qua trinh ldi uu hda sai so tren cac so lieu mau cho

trudc. Bai loan tdi uu hda phi tuyen la mgt bai loan phiic tap, khoi lugng tfnh loan Idn va

noi chung ta khdng cd thudt loan dl xdc dinh dugc nghiem loi uu loan cue.

Trong khudn khd bai bdo nay, mdt mo hinh lai phdi hgp ca hai dang udc lugng co ban la

phi tuyen vd tuyen tfnh se dugc dl xudt va dp dgng.

So dd khdi co bdn cua md hinh lai dugc trinh bay tren hinh la. Tren md hinh nay, so lieu

ddu vao la cdc so lieu qua khir (cdc phu tdi ciiu cac ngay tmdc dd, cac so lieu vl nhiet do trung

binh ciia cdc ngay trudc do). Tin hieu diu ra dugc udc lugng la ldng cua hai thanh phan:

tuydn tfnh va phi tuyln. Thanh phin tuydn tfnh diroc xac dinh bdng thudt loan SVD (Singular

Value Decompositiem - Phan tfch ma Iran theo cac gia tri ky di) [1]. Thanh phan phi tuyln

dirge xac dinh thdng qua vide sir dung md hinh mang MLP (MultiLayer Perceptrem - Mang

298