FE-GRN Neural Network in Hard Disk Failure to Optimize Detection for Enhanced Reliability

As data centers continue to centralize and grow in scale, managing the increasing number of hard disk drives (HDDs) presents significant challenges, necessitating proactive failure prediction. Existing deep learning models for HDD failure prediction primarily focus on long-term sample data, often exhibiting suboptimal performance when applied to short-term sample data. To address this limitation, this paper introduces a novel neural network model called the Feature-Enhanced Generative Recurrent Network (FE-GRN), specifically designed for HDD failure prediction using short-term sample data. The FE-GRN model is composed of three key components: Light Gradient Boosting Machine (LGBM), Time Series Generative Adversarial Network (TimeGAN), and Gated Recurrent Unit (GRU). LGBM is employed for feature selection and rapid iteration, improving the efficiency and accuracy of the analysis. TimeGAN is utilized to generate non-stationary time series data, enhancing the model ability to capture the dynamic behavior of HDDs. GRU, with its gating mechanism, effectively captures time-series dependencies, thereby increasing the model sensitivity to impending faults and improving prediction accuracy. To validate the effectiveness of the proposed model, experiments were conducted using the Backblaze 2023 Q3 HDD dataset and the Baidu dataset. The results indicate that, compared to several related methods, FE-GRN not only effectively addresses the challenge of predicting hard disk failure using short-term sample data but also demonstrates strong generalization capability and robustness.