基于TCBN组合模型的硬盘健康度预测

Hard disk health prediction based on TCBN combined model

针对数据中心硬盘故障单个预测模型性能差且预测结果单一的问题,设计一种基于树型组合贝叶斯网络(TCBN)组合模型的硬盘健康度预测模型。训练4个单一分类器并将输出作为特征节点,提出“SMART-Degree”硬盘健康度划分,将统计后的健康度等级作为贝叶斯网络的分类节点,以全连接方式进行组合得到TCBN组合模型。实验结果表明,模型在故障预测方面真阳率提升至0.857且能较好拟合实际剩余使用寿命。与传统等分健康度划分相比,准确率显著提升,为数据中心可靠性研究提供了一种方案。

Aiming at the problem that a single prediction model’s performance for hard disk failures in data centers is poor and the prediction result is single,a hard disk health prediction model based on the TCBN(tree-type combination Bayesian network)combined model was designed.Four single classifiers were trained and the output was used as the feature node.The SMARTDegree hard disk health division was proposed.The health level was counted as the classification node for the Bayesian network,and it was combined in a fully connected way to obtain TCBN model.Experimental results show that the true positive rate of the model in fault prediction increases to 0.857and it can better fit the actual remaining useful life of the hard disk.Compared with traditional equal health degree division,the accuracy rate is significantly improved,which provides a solution for data center reliability research.