Hard disk drive failure prediction model based on Temporal Convolutional Network combined with Auto-Encoder

Hard disk drive failures pose a significant risk to the normal operation of data centers, making proactive failure prediction crucial for timely issue identification and minimizing potential losses. To effectively predict disk failures, a hard disk drive failure prediction model named AE-TCN is proposed, which combines Auto-Encoder (AE) with Temporal Convolutional Network(TCN) to achieve outstanding predictive performance. First, an improved feature normalization method is employed to select appropriate input features. Then, the Auto-Encoder is used to reduce noise in the input data, while the Temporal Convolutional Network leverages its long-term memory capability for failure prediction. Experimental results demonstrate that, compared to using TCN alone, AE-TCN improves the average Failure Detection Rate (FDR) by 3.317, and the average F1-Score by 2.55% across different time input windows. Moreover, it exhibits better predictive capability compared to other state-of-the-art models.