A Disk Failure Prediction Algorithm Based on Fusion Model.

Disk failure prediction is of great significance for ensuring the business continuity and data reliability of cloud data centers. Algorithm performance will directly affect operation and maintenance costs and user experience. Nowadays, many disk manufacturers use the threshold of S.M.A.R.T. data to check disk fault detection or make prediction, but the detection accuracy of this method is low. This paper proposes a disk fault prediction method based on deep learning, using deep residual network to extract high-dimensional hidden features; and using deep neural network combined with integrated tree model tool XGBoost to perform Failure prediction. Moreover, we use AutoML strategy based on improved GA to get the best model structure. Experiments based on Backblaze data show that this method has better prediction performance. Compared with pure XGboost methods, it can better reduce the false alarm rate about 50% and improve the accuracy rate, while maintaining a strong generalization ability.