A novel prediction model for ship fuel consumption considering shipping data privacy: An XGBoost-IGWO-LSTM-based personalized federated learning approach

Abstract

Due to the protection of shipping data privacy, shipping companies rarely share shipping raw data. Therefore, how to protect shipping data privacy is a crucial issue when predicting ship fuel consumption (SFC). To tackle such issues and guarantee prediction accuracy, a tailored personalized federated learning (PFL) approach is designed and proposed in this study. In the proposed PFL approach, an improved long short-term memory (LSTM) neural network is adopted as the Deep Learning (DL) model for each ship client, in which the improved Grey Wolf Optimization (IGWO) algorithm is juxtaposed to optimize the hyperparameter tuning process in LSTM neural network. In addition, considering that different ships have unique navigation environments and distinct ship characteristics, XGBoost is introduced to make feature selections before LSTM training. Given the personalized features of different ships after XGBoost selection, an FL with personalized layers (Fedper) is employed, where the personalized layers are trained on the client side respectively and the basic layers are uploaded to the central server. To validate the efficiency of the proposed approach, we collect datasets and conduct experiments on 18 bulk carriers. To further prove the superiority of XGBoost-IGWO-LSTM-based PFL, four kinds of comparative experiments are conducted. The results show that our proposed approach can make more accurate SFC prediction results while guaranteeing shipping data privacy. The research could provide meaningful suggestions for shipping companies that prioritize corporate data privacy to reduce SFC.