ABSTRACT

The goal of this paper is to present a denoising approach that can be applied to reduce the measurement noise in model test data used in the identification of the mathematical model for ship manoeuvring motion. To this end' wavelet thresholding based denoising technique is proposed with an intention to smooth out the noise contained in model test data. Support vector regression (SVR) has been wildly used for identifying hydrodynamic derivatives. SVR has a limited allowance for noise in processing polluted test data set directly due to its nature. Wavelet thresholding based denoising techniques are optimal or near optimal for function recovery and can be carried out in three steps: wavelet decomposition' threshold selection and reconstruction. To verify the wavelet denoising in facilitating SVR identification method' the validation tests are carried out based on the simulated data of a KVLCC2 tanker using MMG model. Artificially polluted test data are synthesized with Gaussian white noise. The test shows that'

  1. SVR are highly sensitive to noise.

  2. Wavelet thresholding based denoising can help with SVR performance.

  3. The choice of wavelet thresholding methods doesn't affect the performance of SVR' but the number of wavelet decomposition layers do.

INTRODUCTION

The capability in creating highly accurate mathematical models for simulation of the maneuvering motion is always of great desire and practical value for academic and industry. The study can be traced back in several decades. (Abkowitz 1964) developed whole ship mathematical model to represent hydrodynamic forces/moments with polynomial expression using Taylor expansion. The Mathematical Model Group of Society of Naval Architects of Japan introduced a modular model (Ogawa & Kasai 1978)' which explicitly includes the individual open water characteristics of the hull/propeller/rudder and their interactions. The crux is identifying the unknown hydrodynamic derivatives. System identification (SI) using free trial data has drawn a lot of attention as the advancing development of artificial intelligence. One of the advantages of SI using free run trail is that all hydrodynamic derivatives can estimated with a few free run trial as opposed to numerous captive test (Araki et al. 2012). Support Vector Regression have been proved to be used in system identification (Zhang & Zou 2011; Luo & Zuo; Jian et al. 2015)

This content is only available via PDF.
You can access this article if you purchase or spend a download.