Intelligent Ship Seakeeping Prediction Using Physics-Informed Neural Networks

Seakeeping performance plays a critical role in the safety, efficiency, and comfort of ship operations under complex sea states. Traditional prediction methods based on Computational Fluid Dynamics (CFD) and Response Amplitude Operator (RAO) analysis often suffer from high computational cost and limited adaptability to nonlinear coupled motions. To address these limitations, this paper proposes a physics-informed neural network (PINNs) based intelligent prediction framework for ship seakeeping performance. The method integrates CFD simulations and RAO data to construct a hybrid training dataset and embeds the ship's dynamic response equations derived from energy conservation principles into the loss function of the neural network. A dual-optimizer strategy combining Adam and LBFGS is adopted to ensure both convergence speed and prediction accuracy. Experimental results show that the proposed method achieves high-precision predictions for heave, pitch, and roll responses under various wave conditions, with mean absolute errors below 5.1%. Compared with traditional data-driven or semi-empirical approaches, the PINNs model demonstrates superior generalization ability and physical consistency. The system is further validated in real-case applications, including intelligent navigation risk alert and ship design optimization, showcasing its potential for real-time deployment in intelligent maritime systems.