摘要: |
硬盘广泛应用于各类信息系统中,其工作状态预测对信息系统的正常运行管理有着重要意义。现有基于SMART(Self Monitoring Analysis and Reporting Technology)属性的机器学习预测算法为保证其通用性,普遍选取部分典型属性作为特征,带来一定的信息丢失。在分析SMART数据特点的基础上,提出数据模式分类后再进行机器学习预测的SMART数据处理方法。实际测试结果表明,经分类处理后,采用简单的机器学习算法即可获得与强分类器接近的性能,同时,该方法可有效简化SMART数据机器学习时的特征选择过程,有效降低算法的资源消耗。 |
关键词: HDD硬盘 状态预测 SMART数据模式 机器学习 |
DOI:10.20079/j.issn.1001-893x.220725001 |
|
基金项目: |
|
HDD State Prediction Based on SMART Data Modes |
WAN Chengwei,WANG Xia,WANG Meng |
(Beijing Aerospace Control Center,Beijing 100094,China) |
Abstract: |
The disk is widely adopted in different information systems,and its state prediction is significant for the daily operation and management of those systems.Present self monitoring analysis and reporting technology(SMART)-based machine learning methods usually chose some typical SMART attributes as features for the prediction to ensure the generality.However,some information is lost by this way.A hard drive disk(HDD) state prediction method based on SMART data modes is proposed according to its characteristics,which classifies the SMART records into different subsets according to the data modes.Then,The general machine learning methods are evaluated for these subsets.The test result shows that the proposed method provides the similar performance compared with some strong predictors.At the same time,the feature selection is effectively simplified,and the resource consumption is effectively decreased. |
Key words: hard drive disk state prediction data mode of SMART records machine learning |