Ship pose prediction through an integrated empirical mode decomposition and deep learning method

作者：Zhaozhuo Wang, Gang Yang, Baoren Li, Yue Xu, Chang Yuan, Rui Xu · 发表于：Ships and Offshore Structures · 年份：2025 · DOI：10.1080/17445302.2025.2581075 · 研究领域：Ship Hydrodynamics and Maneuverability、Machine Fault Diagnosis Techniques、Structural Health Monitoring Techniques

Ships operating in complex seas exhibit pronounced six-degree-of-freedom motions—roll, pitch, yaw, heave, sway, and surge—complicating helicopter operations, search and rescue, and unmanned surface vessel autonomy. We construct a realistic prediction setting by synthesizing wave excitations from stochastic models driven by spectral density functions and coupling them with a ship response model to create a 6-DoF dataset. We then propose an attitude prediction framework that integrates empirical mode decomposition (EMD) with deep learning. EMD performs multi-scale decomposition and reconstruction to separate primary trends from high-frequency disturbances, simplifying the signals. A convolutional neural network extracts discriminative features from each intrinsic mode, while a long short-term memory network models temporal dependencies and forecasts future values. The predicted modal components are recombined to yield complete pose predictions. Experiments demonstrate high accuracy and strong generalization across motion modes, indicating the method’s potential to enhance operational safety and autonomy in maritime applications.