Investigation of training data selection in the black-box modeling of ship maneuvering motion

For the identification modeling of ship maneuvering motion, comparisons between various training data are conducted to select appropriate excitation signal with maximum dynamic information, thereby ensuring the generalization ability of the identified model. The identification framework is black-box modeling based on the (“nu”)-support vector machine algorithm with radial basis function kernel, which automatically controls the number of support vectors and keeps sparsity. A Mariner class ship is taken as the study object, and the training data is generated from the reliable simulation model, including 10º/10º, 20º/20º, 30º/30º zigzag maneuvers and 35º turning circle maneuver. The generalization performance of the identified model under different training data is compared by predicting other standard zigzag and turning maneuvers. The results indicate that the 20º/20º and 30º/30º zigzag maneuvers contain more dynamic information and can be used to train the model when the data is pure. The present work provides guidance for the subsequent experiment research to update the ship model quickly in the field.

Subjects

Identification modelling

Multiple excitation signals

Ship manoeuvring

Support vector machine

DDC Class

600: Technik

620: Ingenieurwissenschaften

More Funding Information

National Natural Science Foundations of China

CSSC Joint Fund Project 2017

China Scholarship Council (CSC)

Lizenz

http://rightsstatements.org/vocab/InC/1.0/

Name

paper_0008_file_0454_Investigation_of_Training_Data_Selection_in_the_Black_box_Modeli.pdf

Size

1.38 MB

Format

Adobe PDF

Options

Investigation of training data selection in the black-box modeling of ship maneuvering motion