By harnessing wind energy via wing-sails for supplementary propulsion, wing-diesel hybrid ships achieve significant fuel savings and carbon emission reductions, but their energy efficiency highly depends on collaboratively optimizing navigation paths, wing-sail parameters, and diesel engine output under dynamic meteorological conditions, especially the orientation coupling between wing-sail attack angle, relative wind direction, and ship heading. To address the precision and robustness limitations of traditional reinforcement learning in global navigation energy efficiency optimization, this paper proposes an orientation-aware collaborative optimization framework for energy efficiency of wing-diesel hybrid ships, based on an Adversarial Robust Reinforcement Learning with Transformer (Transformer-ARRL) approach. Firstly, the framework employs an orientation-aware Transformer to extract spatiotemporal features from meteorological and ship operational data for capturing long-term navigation dependencies. Additionally, ARRL is introduced comprising policy, adversary, and critic models, to quantify environmental disturbances for robustness and integrates a module to enhance resilience against environmental uncertainties. Finally, a fuel consumption model based on adaptive boosting algorithm is constructed, incorporating dynamic meteorological data and wing-sail parameters to improve energy efficiency prediction accuracy. Experimental results show that, fundamentally different from other optimization strategies by incorporating orientation-aware perception and adversarial robustness, the Transformer-ARRL approach improves the orientation-coupling efficiency and robustness against environmental perturbations, which ultimately leads to the optimized route achieving 7.14% lower fuel consumption, 255.78-ton fewer CO₂ emissions and 8.31% improvement in the Energy Efficiency Operational Index (EEOI) compared with the initial route.
Wang et al. (Mon,) studied this question.