Dự báo trên chuỗi thời gian sử dụng bài toán tìm kiếm tương tự

Tgp Chi Khoa Hoe Gido Due Ky Thugt (32/2015)

Trudng Bgi Hpc Su Phgm Ky Thuat TR Hd ChiMinh

DU'BAOTRENCHUOlTHdl C3ANStj'DUNGBAITOAN TIMKIEMTl^^

PREDICTION EST TEVIE SERIES USING SIMILARITY SEARCH PROBLEM

Nguyen Thanh Son

Tru&ng dgi hoc Suphgm Ky thudt TP.HCM

l^y toa scan nhan duoc ba/17/3/2015, ngay phan bien <^nh gia (X3/4/2015, ngay ch^ nhan dang 15/4/2015

TOM TAT

Bdi todn du bdo tren chuoi thdi gian Id bdi todn quan trgng trong nhiiu linh vuc vd da nhdn

dugc nhieu su quan tdm tu cdc nhd nghien ciiu trong nhimg ndm gdn ddy. Trong bdi bdo ndy,

chung toi nghien ciru each sd dung bdi todn tim kiem tucmg tu vdo bdi todn du bdo tren chudi

thdi gian^ co xu hu&ng hoge theo miia. Phucmg phdp ndy dirge thuc hiin nhu sau: (I) Trich

mgt chuoi gia tri trin chuoi thdi gian ngay trudc khodng th&i gian mudn du bdo, (2) Sd dung

chuoi ndy de tim k lan can gdn nhdt (hoge cdc lan can trong phgm vi mgt nguong tucmg tu T

chotru&c) dta no trong du lieu qua khu, (3) Trich cdc chuSi (co chiiu ddi bdng vdi chieu ddi

muon du bdo) ngay liin sau moi chuoi lan can tim dugc. vd (4) Chudi du bdo dugc xdc dinh

bdng each tinh trung binh cdc chuoi tim dugc trong budc (3). Kit qud thi^ nghiem cho thdy

edch tiep can ndy cho kit qud (ve do chinh xdc vd th&i gian thuc thi) co thi cgnh tranh dugc

khi so sdnh vdi kit qud du bdo tren chuoi th&i gian co xu hu&ng hoge theo mua su dung mgng

ncr ron nhdn tgo (ANN). Trong thuc nghiem, chiing toi ciing xem xet dnh hu&ng cua kvdT din

do chinh xdc cua du bdo.

Tir khda: Chuoi th&i gian, die bdo, tim kiem tuang tu.

ABSTRACT

Time series forecasting problem is very important pmblem in several domains and has

received a lot of interest from researchers in recent years. In this paper, we investigate the

use of pattern matching technique in seasonal or trend time series prediction. This method

is performed as follows: (1) This technique retrieves the sequence prior to the interval to be

forecasted, (2) This sequence is used as a sample for searching k-nearest neighbors or neighbors

within a threshold Tin historical data, (3) Sequences next to these found patterns are retrieved

(the length of them are equal to the prediction interval), and (4) The forecasted sequence is

calculated by averaging the sequences found in the 3"^ step. The experimental results showed

that this approach produces competitive results on seasonal or trend time series in comparison

to artificial neural network (ANN) in terms of prediction accuracy and time efficiency. In our

experiment, we also examine the impact of parameter values kand Ton the predictive accuracy.

Keywords: time series, prediction, similarity search.

I. GICa THIEU

Mdt chuoi thdi gian la mdt chudi cac sd trong khai pha dii lieu chudi thdi gian. He

thue. Mdi sd bi8u diln mdt gia tri do dugc tai thdng du bao chudi thdi gian du bao cac gia tri

nhiing khoang thdi gian bdng nhau. Dii lieu tuong lai ciia chudi thdi gian bang each xem

chudi thdi gian tdn tai trong nhi^u ling dyng xet dii lieu thu thap dugc trong qua khii. Dp

cua cac ITnh vyc khae nhau nhu khoa hpc, ky chinh xac cua dy bao hen chudi thdi gian se la

thuat, kinh te, tai chinh, y hpe, quan ly hanh co sd cho nhieu tien trinh ra quyet dinh va vi

chinh, v.v.... vay viee nghien cOru cai hen dp hieu qua cua

Du bao tren chudi thdi gian la mdt trong ^^^ phuong phap dy bao se khdng bao gid ket

nhiing cdng viee thach thirc va phiic tap nhdt ^^^^- ^^^ phuong phap du bao thudng dugc