Predicting Ship Main Engine Fuel Consumption Based on Multi-Level Attention Mechanism

This paper integrates the theoretical models of Transformer and BiGRU to construct the Transformer BiGRU Global Attention model, with the aim of enhancing the model’s ability to extract key information. Through the implementation of a cross-attention mechanism to amalgamate features and enhance feature representation, the model attains exact prediction of main engine fuel consumption for vessels. Compared to the Transformer and BiGRU models, our model achieves 86% higher prediction accuracy, enabling more accurate prediction of ship main engine fuel consumption. This furnishes data support for the purpose of comparison with original factory data, thereby facilitating the assessment of engine fault conditions.