采用数据降维的固态硬盘故障检测方法  

A Solid-State Drive Fault Detection Method with Dimensionality Reduction

作　　者：王宇菲 董小社[1] 王龙翔[1] 陈维多 陈衡[1] WANG Yufei;DONG Xiaoshe;WANG Longxiang;CHEN Weiduo;CHEN Heng(Faculty of Electronic and Information Engineering,Xi’an Jiaotong University,Xi’an 710049,China)

机构地区：[1]西安交通大学电子与信息学部,西安710049

出　　处：《西安交通大学学报》2022年第11期176-185,共10页Journal of Xi'an Jiaotong University

基　　金：国家重点研发计划资助项目(2016YFB1000303);国家自然科学基金资助项目(61972311)。

摘　　要：针对固态硬盘SMART数据包含大量高维数据特征导致故障检测准确率不理想的问题,结合固态硬盘SMART数据的时序特点,基于门控循环单元对传统自动编码器的结构做了改进,提出了一种基于门控循环单元稀疏自动编码器降维的固态硬盘故障检测方法(GAL)。首先利用固态硬盘SMART数据训练GRUAE模型,然后利用GRUAE模型中的编码器作为降维的工具,对固态硬盘的原始高维SMART数据进行降维,减少固态硬盘原始SAMRT数据中噪声特征的影响并突出与数据特点更加相关的特征,以提高故障检测的准确率,最后基于降维过的低维SMART数据利用长短时记忆网络进行故障检测。选取阿里巴巴固态硬盘数据集,对多种人工智能算法在准确率、召回率和F_(0.5)方面进行了比较。实验结果表明:相比于没有采用任何降维手段,采用GAL方法可使两种闪存类型的固态硬盘的故障检测准确率、召回率和F_(0.5)分别提高4%、5%、4%和4%、8%、5%,分别达到97%、95%、97%和97%、96%、97%;GAL方法的故障检测准确率、召回率和F_(0.5)分别超出WEFR对比方法53%、25%、43%。The massive high-dimensional SMART data attributes tend to cause a low precision in the detection of solid-state drive(SSD)faults.For this reason,this paper proposes an SSD fault detection method-GAL based on dimensionality reduction with GRU sparse auto-encoder based on an improved structure of the auto-encoder with gated recurrent unit(GRU)according to temporal characteristics of SSD SMART data.First,the GRUAE model is trained with SSD SMART data.Then the encoder of GRUAE model is used as the dimensionality reduction tool to reduce the original high-dimensional SSD SMART data.This is to reduce the influence of noise features in original SSD SAMRT data and highlight the features more relevant to data characteristics to improve the precision of fault detection.Finally,long short-term memory is used for fault detection with low-dimensional SSD SMART data.The precision,recall and F_(0.5) score of various algorithms are compared based on the dataset from Alibaba.The experimental results show that GAL improves the fault detection precision,recall and F_(0.5)-score of the two types of flash SSDs by 4%,5%,4%and 4%,8%and 5%,respectively,reaching 97%,95%,97%and 97%,96%,97%,respectively,compared with the condition with no dimensionality reduction.In addition,the precision,recall and F_(0.5) score of GAL are respectively 53%,25%and 43%higher than those of the method compared.

关 键 词：固态硬盘 故障检测 降维 稀疏自动编码器 门控循环单元 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]