ABSTRACT Purely physics‐based or data‐driven models struggle with the nonlinear, coupled nature of ship propulsion systems, leading to limited accuracy and poor robustness in motion and emission prediction when subjected to realistic and variable operating conditions. To address these limitations, this paper proposes a physics‐informed neural network (PINN) framework as a gray‐box solution that integrates both the white‐box and black‐box modeling paradigms. The proposed hybrid architecture combines the maneuvering modeling group (MMG) model to provide physically consistent constraints and a residual attention neural network trained with observational data to capture residual uncertain nonlinear dynamics. A composite loss function, consisting of data loss and physics‐based loss derived from MMG dynamics, is adaptively weighted during training and backpropagated to optimize network parameters. Experimental results in the SIMMAN2008 data set demonstrate the effectiveness of the proposed approach in predicting both ship motion states and specific fuel oil consumption (SFOC), with SFOC prediction errors of approximately 1.18%. The model also exhibits strong robustness under noise perturbations, offering a reliable, stable, and interpretable solution for the estimation of full‐process ship emissions to support maritime decarbonization.
Han et al. (Thu,) studied this question.