云环境下基于动态滑动窗口多通道Bi-LSTM的虚拟机故障预测模型

针对点值预测方法预测虚拟机故障，未充分利用虚拟机历史周期特征和上下文信息、预测准确率不高的问题，提出了一种动态滑动窗口多通道Bi-LSTM的虚拟机故障预测模型。该模型首先利用动态滑动窗口动态捕获虚拟机故障发生过程的上下文特征；然后构建多通道机制的Bi-LSTM以同时学习不同指标类之间的相关性特征，预测虚拟机下一周期的故障；最后根据OCSVM和区间偏移度方法对预测结果进行判断，得出具体的故障类型。实验表明，该模型在预测准确率、召回率、F值三个指标上均优于基线模型，验证了模型对虚拟机故障预测的有效性。

Virtual machine fault prediction model based on dynamic sliding window multi-channel Bi-LSTM in cloud environment

To address the issue that the point-value prediction method does not make full use of the historical cycle features and contextual information of VM, and the prediction accuracy is not high, this paper proposed a dynamic sliding window multi-channel Bi-LSTM for VM fault prediction. The model firstly used dynamic sliding windows to dynamically capture the contextual features of the VM failure process; then constructed a multi-channel Bi-LSTM to simultaneously learn the correlation features between different indicator classes to predict the VM failure in the next cycle; finally judged the prediction results according to OCSVM and interval skewness methods to derive specific failure types. Experiments show that the model outperforms the baseline model in three metrics, namely prediction accuracy, recall and F-value, verifying the effectiveness of the model for VM fault prediction.