Construction of artificial neural network-driven ship electrical system fault prediction model

This study develops a deep learning-based ship fault prediction model using large-scale sensor data analysis. The model achieves 95.1% accuracy on the test set with a 2.7-hour average early warning time. Transfer learning techniques successfully adapt the model to different ship types, improving the AUC on passenger ship data from 0.823 to 0.912. Five-fold time series cross-validation confirms the model's stability and reliability. The research provides valuable support for ship maintenance decisions and identifies future optimization areas, including electrical system fault prediction and performance under extreme weather conditions.