Computational performance benchmarking.

High-Voltage Direct Current (HVDC) transmission systems require fast and reliable fault diagnosis to ensure secure and stable operation. However, existing methods, including conventional Convolutional Neural Networks (CNNs), often suffer from limited accuracy and degraded training performance as network depth increases. To address these limitations, this study proposes an improved one-dimensional Residual Neural Network (1D-ResNet) that integrates an attention mechanism within the residual blocks to enhance feature extraction, stabilize gradient propagation, and accelerate model convergence. A comprehensive simulated HVDC platform is established to generate multiple fault scenarios, and the proposed network is trained to identify one normal condition and six typical fault types. Experimental results demonstrate that the proposed method achieves an average diagnostic accuracy of 99.15%, outperforming traditional CNN-based approaches by 12.89%. Moreover, the loss value is significantly lower than that of the conventional CNN model, indicating substantial improvements in both robustness and learning efficiency. These findings confirm the effectiveness of the proposed attention-enhanced residual framework for high-precision HVDC fault diagnosis.