基于AE-LSTM的多目标硬盘故障预测方法

A Multi-objective HDD Failure Prediction Method Based On AE-LSTM

摘要:

硬盘故障预测是在故障发生前发出预警,避免数据丢失或服务中断,提高数据中心的可靠性和安全性。然而,大多数故障预测模型将硬盘故障问题转化为二分类任务,忽略了硬盘故障是渐变过程的,并且缺乏故障诊断功能。因此,提出了一种基于AE-LSTM的硬盘故障预测框架,实现多目标任务：硬盘健康状态分级、硬盘剩余使用寿命预测、硬盘故障诊断。首先,采用回归决策树模型智能化对硬盘健康状态进行标记；其次,通过AE-LSTM模型提取鲁棒的隐藏变量,并构建剩余使用寿命预测模型和硬盘健康状态分级模块；最后,根据AE模块的输入输出差异进行硬盘故障诊断。在Backblaze公开数据集上,对比了RF、LSTM和AE-LSTM三种算法,实验结果证实了AE-LSTM算法在多目标硬盘故障预测中的有效性和优势。

Abstract:

HDD failure prediction can be used to protect data loss or service interruption by early warning before real failure of HDD, and improve the reliability and security of data center. However, most studies focused on HDD failure prediction as binary classification and ignored the gradual deterioration of HDD while lacking fault diagnosis function. Therefore, this paper proposes a HDD failure prediction method based on AE-LSTM to achieve multi-objective tasks: HDD health status multi-classification, HDD RUL prediction and HDD fault diagnosis. Firstly, the decision tree regression was used to intelligently label HDD health status. Secondly, the robust hidden variables were extracted by AE-LSTM from SMART, fed to RUL model and HDD health status classification model. Finally, the HDD fault diagnosis is implemented by computing the difference between the input and output of AE module. By evaluating RF, LSTM and AE-LSTM algorithms on the Backblaze public dataset, experimental results showed that the effectiveness and advantages of AE-LSTM algorithm in multi-objective HDD failure prediction.