A Failure Prediction Approach Based on BiLSTM and Deep Feature Extractor for Hard Disk Drives
Dongjiang Li, Wenyuan Qiu, Jing Zhang, Kun Xue, Xianbo Zhang, Binbin Sun, Feng Lin, Li Li · 2023
With the swift implementation of cloud platforms, guaranteeing optimal service dependability is imperative. Disk failure is a common cause of service unreliability, particularly in industrial cloud platforms that house numerous disks. These approaches can forecast disk failures by analyzing disk status data before they occur. Deep neural networks have demonstrated their effectiveness in addressing time series classification problems. Our research paper introduces an innovative and efficient approach to forecasting disk failures. By incorporating Bidirectional Long Short Term Memory (BiLSTM) into deep neural networks, we enhance the ability to extract relevant features. We explore the use of very neural networks with the ResNet structure and Fully Convolutional Network. The attention mechanism is adopted to support BiLSTM in acquiring long-term dependencies within lengthy sequences. Our model’s effectiveness is evaluated on two datasets and compared to advanced approaches. The empirical findings unequivocally establish that our performance surpasses that of alternative techniques for predicting disk failures.