Research on Ship Fuel Consumption Prediction Based on LSTM

This article proposes a study on fuel consumption modeling and prediction based on data obtained from actual navigation monitoring of inland vessels, utilizing LSTM neural networks for modeling. The method selects 12 fuel consumption-related parameters, including main engine power and torque. Through analyzing the correlation and importance of the ship's fuel consumption characteristics, the significance of each factor's impact on fuel consumption is determined, verifying the interpretability and accuracy of the model. Finally, by comparing evaluation metrics with four other algorithms SVM, RFG, XGB, and GTB it was found that the LSTM model reduces the MSE by an average of 43.2%, MAE by 27.9%, and MAPE by 76.8%. This indicates that the LSTM neural network model has higher predictive accuracy, interpretability, and generalization ability compared to other prediction models, achieving a more precise fit to the target data.