A Novel ROA-Optimized CNN-BiGRU Hybrid Network with an Attention Mechanism for Ship Fuel Consumption Prediction

Optimizing ship energy efficiency and advancing the green transition of the shipping industry depend on an accurate model for predicting ship fuel consumption (FC). This study builds a hybrid prediction model that combines a Convolutional Neural Network (CNN), Bidirectional Gated Recurrent Unit (BiGRU), and an attention mechanism using operational data from ships. The model is tuned using the Red Kite Optimization Algorithm (ROA). First, correlations between ship navigational environmental data and operational data are analyzed, and cluster analysis is performed to select suitable input features. Subsequently, the ship FC prediction model based on ROA-CNN-BiGRU-Attention (RCGA) is developed. A case study shows that the RCGA model reaches a root mean square error (RMSE) as low as 0.0205 and an R2 value as high as 0.9330, demonstrating strong performance in dynamic shipping scenarios, with advantages in handling temporal dependencies and complex operational patterns. Moreover, the model exhibits reasonable robustness, providing some support for ship energy efficiency optimization and assisting the shipping industry in advancing low-carbon development and sustainable green transition.