基于变权重随机森林的硬盘故障预测方法

Hard disk failure prediction method based on variable weight random forest

为保证一定硬盘故障预测准确率并消除预测误判率,提出一种基于变权重随机森林模型的优化方法。采用计算特征属性值和硬盘故障的相关性对原始数据集进行降维处理,考虑到决策树的节点分裂信息值可能为0的情况,提出分裂信息值与分裂信息平均值之和来代替单一的分裂信息值;根据精度和多样性值选取较优决策树,为其动态分配权值组成强分类器随机森林模型。该模型在保证极高故障预测准确率的同时,将故障预测误判率降低到了0.008%。相比较之前的模型,准确率提高的同时误判率低至0,为解决预测硬盘故障的问题提供了一种借鉴思路。

To ensure the accuracy of hard disk fault prediction and eliminate the miscalculation rate of prediction,an optimization method based on variable weight random forest model was proposed.The dimension reduction of the original data set was carried out by calculating the correlation between the characteristic attribute value and the hard disk fault.Considering that the node splitting information value of the decision tree may be 0,the sum of the splitting information value and the splitting information average value was proposed to replace the single splitting information value.According to the precision and diversity,the optimal decision tree was selected and the dynamic distribution weight was assigned to form a strong classifier random forest model.This model not only ensures the high accuracy of fault prediction,but also reduces the error rate of fault prediction to 0.008%.Compared with the previous model,the accuracy is improved while the misjudgment rate is as low as 0,which provides an idea for solving the problem of predicting hard disk faults.