Date of Award

5-2026

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Ocean Engineering and Marine Sciences

First Advisor

J. Travis Hunsucker

Second Advisor

Seong Hyeon Hong

Third Advisor

Stephen Wood

Fourth Advisor

Richard B. Aronson

Abstract

This work proposes a residual hybrid MMG-GP model to address the growing trend in marine autonomy requiring ship maneuvering predictions that are both accurate and reliable across a range of operational conditions. The model combines a Maneuvering Modeling Group (MMG) model with a Gaussian Process (GP) corrective term to combine the strengths of both physics and data-driven methods. The hybrid MMG-GP model and the individual MMG and equivalent black-box GP model are compared using positional and energy error metrics for a 35° turning circle and a 10°/10° zigzag maneuver. Training data are varied in duration and rudder excitation frequency to examine the effects of common system identification limitations across 3200 independent samples. The MMG-GP demonstrated consistent improvements over the MMG model and achieved an average error reduction of 58.2% compared to its black-box counterpart, with larger gains when trained on limited and low-excitation data. Results indicate that a physics model corrected via machine learning is a simple yet effective method for improving maneuvering modeling applications.

Available for download on Monday, November 09, 2026

Share

COinS