A Ship Fuel Consumption Prediction Method Based on the Transformer-Local Attention Model

This study proposes an innovative deep learning model that combines the local attention mechanism with Transformer architecture to achieve high-precision Marine fuel consumption prediction. Firstly, the massive data collected by ship sensors are preprocessed, and the fuel consumption is smoothed through filters to reduce noise interference and improve data quality. The model has unique multi-scale feature processing capabilities: the local attention mechanism accurately captures local features in the ship's operational data, while the Transformer architecture digs deeply into the overall data pattern with powerful global learning capabilities. Through the dynamic weighted fusion strategy, the model adaptively adjusts the importance of global and local features in the training process, thus accurately characterizing the ship's operating state and realizing the fuel consumption prediction based on multi-feature input. Experimental results show that the model exhibits outstanding performance, with a mean squared error (MSE) of only 0.000212. This research provides a high-precision fuel consumption prediction solution for ship energy efficiency management and has important practical significance for energy conservation and emission reduction in the shipping industry. At the same time, it helps optimize ship operation decisions and promote the development of the shipping industry in a green and sustainable direction.