Online Nonparametric Identification Modeling of Intelligent Ship Maneuvering Dynamics Based on Expectation- Maximization Algorithm

Accurate identification of ship maneuvering dynamics is a fundamental prerequisite for realizing autonomous navigation in intelligent ship systems. Existing nonparametric identification methods face critical limitations under realistic data-stream conditions: batch-mode algorithms cannot process streaming sensor data in real time, while parametric approaches impose rigid assumptions on the underlying system structure. This paper proposes an online nonparametric identification framework for intelligent ship maneuvering dynamics based on the Expectation-Maximization (EM) algorithm, specifically, an Online EM (OEM) variant adapted for sequential data streams. The proposed method treats ship maneuvering forces and moments with a probabilistic Gaussian mixture framework and iteratively refines both model parameters and latent structure using incoming sensor observations, without requiring a pre-specified model order. The method is designed to handle the nonlinearity and non-Gaussianity of ship motion under environmental disturbances, including wind and current. Systematic experiments are conducted on the SR108 container ship dataset, encompassing turning tests and zigzag tests. Comparative evaluations against the incremental Gaussian mixture model (IGMM) demonstrate that the proposed OEM-based method achieves superior prediction accuracy and real-time adaptability. The proposed framework provides a computationally efficient and practically deployable solution for online, structure-free modeling of intelligent ship maneuvering systems.