Int J Performability Eng ›› 2019, Vol. 15 ›› Issue (4): 1161-1170.doi: 10.23940/ijpe.19.04.p11.11611170

Previous Articles     Next Articles

Hybrid SVM and ARIMA Model for Failure Time Series Prediction based on EEMD

Haiyan Suna, Jing Wua, *, Ji Wub, *, and Haiyan Yangb   

  1. a School of Mathematics and Systems Science, Beijing University of Aeronautics and Astronautics, Beijing, 100191, China;
    b School of Computer Science and Engineering, Beijing University of Aeronautics and Astronautics, Beijing, 100191, China
  • Revised on ; Accepted on
  • Contact: E-mail address: sunhy@buaa.edu.cn
  • About author:Sun Haiyan is an associate professor of the School of Mathematics and Systems Science at Beijing University of Aeronautics and Astronautics. He received his Ph.D. from Nankai University in 1997, and he worked as a postdoctoral researcher in the Department of Statistics at Renmin University of China from 1997 to 2000. He focused on probability statistics. His research interests include statistical techniques, statistical model diagnosis, and statistical data analysis in software reliability analysis. Wu Jing is currently pursuing her Master’s degree at Beijing University of Aeronautics and Astronautics. She received her B.S. degree from Capital Normal University in 2016. Her research is focused on software reliability analysis based on time series. Wu Ji is an associate professor and assistant dean of the School of Computer Science and Engineering (SCSE) at Beijing University of Aeronautics and Astronautics. He received his Ph.D. from Beijing University of Aeronautics and Astronautics in 2003 and his M.S. degree from the Second Research Institute of the China Aerospace Science and Industry Group in 1999. He focuses on industry-oriented research. His research interests include embedded systems and software modeling and verification, software requirements and architecture modeling and verification, safety and reliability assessment, and software testing. Yang Haiyan is a lecturer at Beijing University of Aeronautics and Astronautics. Her main research interests include software engineering, software safety, and software testing.

Abstract: A more widely used hybrid model of support vector regression (SVR) and autoregressive integrated moving average (ARIMA) based on Ensemble Empirical Mode Decomposition (EEMD) is proposed for failure time series prediction by taking advantage of the SVR model to forecast the nonlinear part of failure time series and the ARIMA model to predict the linear basic part. It firstly uses EEMD to decompose the original failure sequence into several significant fluctuation components and a trend component, and then it utilizes SVR and ARIMA to forecast them separately. The performance of the presented model is measured against other unitary models such as Holt-Winters, autoregressive integrated moving average, multiple linear regression, and group method of data handling of seven published nonlinear non-stationary failure datasets. The comparison results indicate that the proposed model outperforms other techniques and can be utilized as a promising tool for failure data forecast applications.

Key words: ensemble empirical mode decomposition, support vector machines regression, autoregressive integrated moving average, failure time series forecast, hybrid models