Time series online prediction method based on information perception weight and error prediction

被引：0

作者：

Wang, Hao ^{[1
,2
]}

Liu, Zhen ^{[1
]}

机构：

[1] School of Automation Engineering, University of Electronic science and Technology of China, Chengdu,611731, China

[2] Unit 91388 of the PLA, Zhanjiang,524022, China

来源：

Yi Qi Yi Biao Xue Bao/Chinese Journal of Scientific Instrument | 2020年 / 41卷 / 11期

关键词：

Aiming at the problems of insufficient perception of changes in data characteristics and the insufficient timeliness of prediction in existing time series online prediction methods; this article innovatively designs a time series online prediction method based on information perception weight and error prediction. The method uses the information perception weight to replace the forgetting factor parameter λ0 in the cost function; through establishing the mapping relationship between input data and prediction error; error prediction is performed; then error compensation is realized using weighting error compensation coefficient. Multiple single-step prediction experiments were carried out through using the method of changing the number of hidden layer nodes. The experiment results verify the excellent single-step prediction ability of the design method in terms of prediction accuracy and generalization. Among them; the single-step prediction variances of Sinc; Mackey-Glass and Solar Energy that are the three data selection points reach 1.56×10-13; 2.29×10-7; and; 1.43; respectively. According to the actual failure situation; the failure voltage was set to 5.8V and 5.6V; respectively; multiple-step prediction was performed aiming at the actually measured data in the accelerated life test of the packaged step-down power module. The five-step and ten-step prediction results show that the design method can effectively predict power failure. The experiment results fully demonstrate that the design method can complete online single-step and multi-step predictions stably; accurately and effectively when the predicted data characteristics are changed. © 2020; Science Press. All right reserved;

D O I：

暂无

中图分类号：

学科分类号：

摘要：

引用

页码：31 / 41

共 50 条

[1] Support vector machine method for financial time series prediction based on simultaneous error prediction
Zhang, Z. (zhangzs@tju.edu.cn), 1600, Tianjin University (47):
[2] Time Series Prediction Based on Online Learning
Song, Q.
2015 IEEE 14TH INTERNATIONAL CONFERENCE ON MACHINE LEARNING AND APPLICATIONS (ICMLA), 2015, : 857 - 864
[3] ERROR OF PREDICTION OF A TIME SERIES
BLOOMFIELD, P
BIOMETRIKA, 1972, 59 (03) : 501 - 507
[4] NonSTOP: A NonSTationary Online Prediction Method for Time Series
Xie, Christopher
Bijral, Avleen
Ferres, Juan Lavista
IEEE SIGNAL PROCESSING LETTERS, 2018, 25 (10) : 1545 - 1549
[5] Time Series Prediction Based on Adaptive Weight Online Sequential Extreme Learning Machine
Lu, Junjie
Huang, Jinquan
Lu, Feng
APPLIED SCIENCES-BASEL, 2017, 7 (03):
[6] Time Series Prediction Based on Online Sequential Improved Error Minimized Extreme Learning Machine
Xue, Jiao
Liu, Zeshen
Gong, Yong
Pan, Zhisong
PROCEEDINGS OF ELM-2015, VOL 1: THEORY, ALGORITHMS AND APPLICATIONS (I), 2016, 6 : 193 - 209
[7] Estimation of prediction error in time series
Aue, Alexander
Burman, Prabir
BIOMETRIKA, 2024, 111 (02) : 643 - 660
[8] Multivariate Time Series Prediction Based on Temporal Change Information Learning Method
Zheng, Wendong
Hu, Jun
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS, 2023, 34 (10) : 7034 - 7048
[9] A novel method for time series prediction based on error decomposition and nonlinear combination of forecasters
Chen, Wei
Xu, Huilin
Chen, Zhensong
Jiang, Manrui
NEUROCOMPUTING, 2021, 426 (426) : 85 - 103
[10] Discounted online Newton method for time-varying time series prediction
Ding, Dongsheng
Yuan, Jianjun
Jovanovic, Mihailo R.
2021 AMERICAN CONTROL CONFERENCE (ACC), 2021, : 1547 - 1552

← 1 2 3 4 5 →