Research On Data Center Disk Failure Prediction Model

With the advent of the era of big data,the storage of massive data is gradually transferred from the client to the cloud and stored in the enterprise data center.Once the data center fails,it will bring a lot of losses to the enterprise.Nowadays,data storage in modern data centers still dominates disk storage,so the maintenance of disk failure is one of the main tasks of today's operation and maintenance personnel.How to effectively use existing resources to predict future disk failures based on disk data,and timely backup data replacement disks to improve the reliability of modern data center storage systems has become one of the research hotspots of data centers.However,there are still some shortcomings in the current data center disk failure prediction research.At present,the main disk failure prediction modeling methods are processed by classification problems,which can only predict whether the disk will fail in the future,and the current disk failure.The research accuracy of the prediction problem is not high enough,and the effective prediction time of the disk failure prediction model is also short.The operation and maintenance and related personnel cannot timely back up and process the disk according to the results predicted by the disk.Aiming at the above problems,this thesis proposes a disk failure prediction model,which is divided into a disk failure detection model based on convolutional neural network and a SMART data prediction model based on LSTM neural network and fbprophet algorithm,which can accurately predict the next 30 days.Disk failure condition.The main contents are as follows:(1)Disk failure detection model uses convolutional neural network to calculate and utilize local processing data and weight sharing,so that it can automatically and effectively extract useful features of data while calculating high-dimensional input data and calculate it.Characteristics of the purpose of the classification.Feature extraction is performed on the disk information by using multiple convolution layers and a pooling layer to obtain a result of detecting whether the disk is broken and the current disk information,thereby detecting whether the current disk is broken.(2)SMART data prediction model uses LSTM neural network to solve the problem that long-term dependence problem and fbprophet algorithm have better fitting effect on periodic and holiday features of time series data.The two models are combined to improve the prediction accuracy of disk SMART data.The model data fusion model SMART data output model is input into the disk failure detection model to establish a disk failure prediction model that can determine the specific failure time in the future.Experiments show that the model can accurately predict the failure information of the disk in the next 30 days.This thesis provides a reference for the design and implementation of future disk failure prediction systems.