Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data Augmentation Methods of Dynamic Model Identification for Harbor Maneuvers using Feedforward Neural Network

Kouki Wakita, Yoshiki Miyauchi, Youhei Akimoto, Atsuo Maki

TL;DR

Data augmentation is introduced to improve the generalization performance of dynamic models identified from a limited dataset and slicing and jittering are used as data augmentation methods and confirmed their effectiveness by numerical experiments using the free-running model tests.

Abstract

A dynamic model for an automatic berthing and unberthing controller has to estimate harbor maneuvers, which include berthing, unberthing, approach maneuvers to berths, and entering and leaving the port. When the dynamic model is estimated by the system identification, a large number of tests or trials are required to measure the various motions of harbor maneuvers. However, the amount of data that can be obtained is limited due to the high costs and time-consuming nature of full-scale ship trials. In this paper, we improve the generalization performance of the dynamic model for the automatic berthing and unberthing controller by introducing data augmentation. This study used slicing and jittering as data augmentation methods and confirmed their effectiveness by numerical experiments using the free-running model tests. The dynamic model is represented by a neural network-based model in numerical experiments. Results of numerical experiments demonstrated that slicing and jittering are effective data augmentation methods but could not improve generalization performance for extrapolation states of the original dataset.

Data Augmentation Methods of Dynamic Model Identification for Harbor Maneuvers using Feedforward Neural Network

TL;DR

Data augmentation is introduced to improve the generalization performance of dynamic models identified from a limited dataset and slicing and jittering are used as data augmentation methods and confirmed their effectiveness by numerical experiments using the free-running model tests.

Abstract

A dynamic model for an automatic berthing and unberthing controller has to estimate harbor maneuvers, which include berthing, unberthing, approach maneuvers to berths, and entering and leaving the port. When the dynamic model is estimated by the system identification, a large number of tests or trials are required to measure the various motions of harbor maneuvers. However, the amount of data that can be obtained is limited due to the high costs and time-consuming nature of full-scale ship trials. In this paper, we improve the generalization performance of the dynamic model for the automatic berthing and unberthing controller by introducing data augmentation. This study used slicing and jittering as data augmentation methods and confirmed their effectiveness by numerical experiments using the free-running model tests. The dynamic model is represented by a neural network-based model in numerical experiments. Results of numerical experiments demonstrated that slicing and jittering are effective data augmentation methods but could not improve generalization performance for extrapolation states of the original dataset.
Paper Structure (16 sections, 24 equations, 11 figures, 5 tables)

This paper contains 16 sections, 24 equations, 11 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Parameter Identification Methods
  3. Coordinate systems
  4. Optimization problem
  5. Data Augmentation Methods
  6. Definition of the reference dataset
  7. Data augmentation by slicing
  8. Data augmentation by jittering
  9. Data augmentation by slicing and jittering
  10. Numerical Experiments
  11. Dataset
  12. Mathematical Model
  13. Optimization Methods and Results
  14. Prediction Results
  15. Discussion
  16. ...and 1 more sections

Figures (11)

  • Figure 1: Coordinate systems.
  • Figure 2: Subject model ship at the experimental pond.
  • Figure 3: Histograms of ship and actuator state variables. Note that the vertical axes, which show the frequency, are scaled logarithmically.
  • Figure 4: 2D histogram of apparent wind speed and direction. The bin width of apparent wind speed is $0.25$ m/s, and that of apparent wind direction is $20$ degrees. The color bar shows the frequency with a logarithmic scale.
  • Figure 5: Exponential moving average values of the evaluation function in the validation dataset. The legend implies the used training dataset. Ten training results with different random numbers for each dataset are presented.
  • ...and 6 more figures