Data-Driven Modeling of Ship Maneuvering Dynamics via Control-Aware Hankel-DMDc and Mode Attribution

Accurate data-driven identification and long-term forecasting of surface ship motions in realistic ocean environments are essential for advanced autonomous navigation and predictive control systems. However, robust system identification in such conditions remains challenging due to high-frequency wave-induced disturbances, which contaminate measured signals and may contribute to error accumulation in recursive forecasting. This paper proposes a control- and wave-aware data-driven framework for modeling and prediction of ship maneuvering dynamics in stochastic wave environments. The approach integrates Hankel dynamic mode decomposition with control (Hankel-DMDc) with a spectral Mode Attribution Layer that characterizes modal associations with commanded control inputs and measured wave excitation. A dual-criterion selection rule, based on the Wave Energy Projection (WEP) and Control Correlation Index (CCI), is introduced to characterize and separate maneuvering-related and wave-associated modal content without relying on a predefined physical model. The framework is evaluated on a multidimensional ship-motion dataset from the DaRUS repository, using routine maneuvering trajectories for model identification and independent out-of-distribution test trajectories for prediction evaluation. One-step prediction analysis indicates high local accuracy, particularly for rotational states, while recursive multi-step forecasting shows gradual error growth over a 200-step prediction horizon. A quantitative comparison with unfiltered Hankel-DMDc, conventional Butterworth filtering, and direct wave-input modeling further shows that the proposed framework retains predictive performance comparable to direct Hankel-DMDc for the present dataset, while providing explicit modal attribution and wave/maneuvering characterization. The identified discrete-time operator also exhibits a spectral radius below unity, while no rapid divergence is observed over the evaluated recursive prediction horizon. The proposed methodology provides an interpretable data-driven framework for modeling marine systems, with potential applicability to control-oriented system identification, predictive modeling, autonomous navigation, and Model Predictive Control (MPC).