Hard Drive Failure Detection Technology Based on GRU-AE Dimensionality Reduction

The stability of hard drives directly impacts the reliability and continuity of operations at provincial meteorological data centers. Implementing intelligent monitoring of hard drive failures can significantly enhance operational efficiency. This paper employs a GRU-AE model to improve the dimensionality reduction of hard drive status data, transforming high-dimensional SMART attributes into low-dimensional feature representations. This approach reduces the influence of noisy features while highlighting characteristics more relevant to hard drive failures. Subsequently, the dimensionality-reduced data is processed through a deep convolutional neural network to extract predictive features, which are then combined with a Long Short-Term Memory (LSTM) neural network for forecasting. This method addresses the long-term backpropagation challenges across network layers. Experimental results demonstrate that the proposed approach achieves good performance in both accuracy and recall rate for hard drive failure detection tasks.