Predicting steady degradation in ship power system: A deep learning approach based on comprehensive monitoring parameters

Abstract

Abstract Steady degradation (SD) prediction is crucial for the intelligent operation and maintenance of ship power system (SPS). Addressing the challenge of predicting the SD process, this study introduces the YC2Model, a system‐level predictive method that integrates encoding time slice data to images (ETSD2I) with a convolutional neural network and Transformer. Incorporating the Transformer, in particular, enables the YC2Model to predict the SD state of SPS over extended periods more effectively. Compared to baseline models, YC2Model demonstrates superior performance on key performance indicators, including the highest coefficient of determination (R2) of 0.960717, and the lowest symmetric mean absolute percentage error of 0.015500, mean square error of 0.707211 × 10−4, root mean square error of 0.008410, and mean absolute error of 0.006519, proving its superior predictive accuracy. The correlation between model performance variations and degradation mechanisms is validated through statistical analysis of the YC2Model's performance in different stages of the SD process. During the SD process, YC2Model exhibits high predictive accuracy, an ability to capture changes in degradation mechanisms and robust adaptability to degradation trends. This model can provide precise and reliable SD state predictions for the intelligent operation and maintenance of SPS.

Keywords

• comprehensive monitoring parameters
• convolutional neural networks
• ship power system
• steady degradation prediction
• transformer