基于非线性多参数模型的软件老化检测

摘要

This paper presented a method of nonlinear autoregressive models with exogenous inputs to detect the aging phenomenon of the software system. It figures out the problem existing in current software aging methods that there is no cosideration of the correlation between multivariate and the impact of delay from historical data. We first collected the performance data of the HelixServer-VOD server, did principal component analysis of the data,determined the input dimension, determined the best model order according to AIC criteria, selected the reasonable network model structure eventually. We used the known imaging state samples to train the NARX network in order to establish the identification model of the system, then hypothesis tested the residual of the NARX identification model through the method of sequential probability ratio test,finally judge the aging condition of the system. The experimental result shows that NARX model-based fault detection method can be effectively applied in checking software aging.%提出了一种软件系统的非线性有源自回归(Nonlinear AutoRegressive models with eXogenous Inputs,NARX)网络模型的老化检测方法.解决了目前软件老化方法未考虑多变量间关联性及历史数据的延迟影响的问题.该方法首先通过对实验采集的HelixServer-VOD服务器性能数据进行主成分分析,确定网络的输入维数,根据AIC准则确定最佳模型阶数,最终选取合理的网络模型结构；使用已知的未老化状态样本对NARX网络进行训练,建立系统的辨识模型；然后运用序贯概率比检验(Sequential Probability Ratio Test,SPRT)对NARX辨识模型的残差进行假设检验,判断系统的老化状态.实验分析表明,基于NARX网络模型的故障检测方法能够有效地应用于软件老化的检测.