A privacy-preserving federated transfer learning with ship mapping (FTL-SM) framework for accurate ship energy efficiency prediction

作者：Shaohan Wang, Ruihan Wang, Feiyang Ren, Min Chen, Ran Yan · 发表于：Advanced Engineering Informatics · 年份：2025 · DOI：10.1016/j.aei.2025.103569 · 被引用次数：7 · 研究领域：Maritime Transport Emissions and Efficiency、Air Quality Monitoring and Forecasting、Maritime Navigation and Safety

The concern of protecting shipping data privacy restricts the sharing of energy efficiency data among shipping companies, posing challenges for accurate ship fuel consumption (SFC) prediction. To address this concern, this study proposes a federated transfer learning with ship mapping (FTL-SM) framework for SFC prediction, which enhances data consistency in collaborative training, optimizes knowledge transfer, and safeguards data privacy. The FTL-SM framework operates under a federated learning setting, ensuring that only model parameters are exchanged between participants, while raw fuel consumption data remains securely localized. The proposed framework consists of two integrated components: a ship mapping model and a fuel consumption prediction model. First, a random forest-based ship mapping model classifies ships into groups based on static attributes (e.g., ship dimensions, engine configurations, and structural specifications), thereby enhancing data homogeneity and facilitating more effective knowledge transfer. Second, a domain knowledge-informed artificial neural network (DK-ANN) model is employed to predict SFC, explicitly embedding monotonicity and convexity constraints to align model behavior with physical and operational principles. To further safeguard data confidentiality and improve training robustness, the FTL-SM framework incorporates a dynamically adjusted differential privacy mechanism and a proximal regularization mechanism during local model optimization...