Ultra-high-voltage(UHV)submarine cables are a key component of renewable integration and island power grids

and reliable diagnosis of grounding faults is essential for their safe and stable operation. A fault-diagnosis method was proposed for submarine cables based on a hybrid one-dimensional convolutional neural network and bidirectional long short-term memory network(1D-CNN-BiLSTM). The one-dimensional time series of sheath and armor grounding currents measured at both ends were used as inputs. A 1D-CNN first performs automatic extraction of local high-dimensional temporal features with length compression

and a BiLSTM then models forward and backward dependencies to capture feature variations during fault evolution

thereby improving diagnostic accuracy. The proposed model was evaluated on a 500 kV HVAC submarine-cable system between Ningbo and Zhoushan Island. Results show that

compared with commonly used methods

the model achieves an accuracy of 97.66% with an average inference time of 0.059 s per sample; under additive noise

the accuracy remains 97.13%

a decrease of only 0.53% relative to the noise-free case

demonstrating strong noise robustness. By enabling automatic feature extraction and precise identification of typical grounding faults

the proposed approach provides a practical diagnostic solution and engineering reference for fault identification and operation maintenance of UHV AC submarine-cable systems.