Abstract
Nowadays, software systems have been more and more complex, which causes great challenges to maintain the availability of the systems. Online failure prediction provides an effective approach to guaranteeing the validity of the systems. Most of the current technologies for online failure prediction require some prior knowledge, such as the model of the system or failure patterns. This paper proposes a new method based on fuzzy rules and time series analysis. Specifically, fuzzy rules are used to model the relationships among different variables, whereas univariate time series analysis is used to describe the evolution of each variable. Thus, for a dependent variable, we have two predicted values: one is from the time series model, and the other is computed from fuzzy rules with fuzzy inference. If the difference between the two values exceeds a threshold, then we declare that there would be a failure in some time period ahead. Different from the existing methods, the proposed method considers not only the evolutionary trend of each variable but also the relationships among different variables. Moreover, we do not need any prior knowledge such as system model or failure patterns. We use a railway transportation system as an example to illustrate our method.
Original language | English (US) |
---|---|
Article number | 8357501 |
Pages (from-to) | 1143-1158 |
Number of pages | 16 |
Journal | IEEE Transactions on Reliability |
Volume | 67 |
Issue number | 3 |
DOIs | |
State | Published - Sep 2018 |
All Science Journal Classification (ASJC) codes
- Safety, Risk, Reliability and Quality
- Electrical and Electronic Engineering
Keywords
- Autoregressive integrated moving average (ARIMA)
- failure prediction
- fuzzy rules
- railway transportation system
- time series analysis