Dynamic optimization of ship EEOI using a BO-BiLSTM prediction model and annealing evolutionary algorithm

To comply with IMO regulations on ship energy efficiency and carbon intensity index (CII) ratings, this paper proposes a dynamic optimization strategy for the energy efficiency operational indicator (EEOI) using a Bayesian optimization bidirectional long short-term memory network (BO-BiLSTM) prediction model and an annealing evolutionary algorithm (AEA). A bidirectional long short-term memory network (BiLSTM) with Bayesian optimization (BO) is adopted to predict fuel consumption rate due to its capability in capturing bidirectional temporal dependencies and enabling efficient hyperparameter tuning under complex marine operating conditions. The AEA is developed to optimize main engine revolution under operational constraints, forming a dynamic EEOI optimization framework. The results show that BO-BiLSTM significantly outperforms baseline models in prediction accuracy, while the optimized speed scheme reduces EEOI by an average of 9.57% and by 17.69% in segments under adverse sea states. The proposed method provides an effective technical approach for improving ship emission efficiency and supporting compliance with the International Maritime Organization (IMO) energy efficiency standards.