ABSTRACT

It is important to comprehend propulsive performance of a ship in actual seas by means of analyzing on-board monitoring data. In my previous study, I applied Gaussian process regression (GPR), which is one of machine learning, to regress the on-board monitoring data and calculate propulsive performance of a ship against Beaufort scale so as to show a way of the usage results of the regression. The estimator of the GPR shows good agreement with measured time-history data which are used as train data and test data but predicts unreasonable results physically of speed-power relationship, so called speed-power curves against Beaufort scale. In this paper, the author applies Multi-fidelity Gaussian process regression (MF-GPR) to solve the problem. MF-GPR uses on-board monitoring data as high-fidelity data and results of computational simulation based on physical model as low fidelity data. The estimator of the MF-GPR model shows reasonable results physically of speed-power curves.

INTRODUCTION

In the shipping industry, various regulations for emission control, for instance, Energy Efficiency index (EEXI) and Carbon Intensity Indicator (CII) have been introduced for ship performance in recent years (IMO, 2009; IMO, 2012; IMO, 2021). In order to comply with these regulations, it is very important to comprehend ship propulsive performance in actual operation load conditions in various weather conditions. It is expected that analyzing on-board monitoring data, which include voyage data record (VDR) or engine data logger contributes to comprehend the ship propulsive performance. In this paper, the author obtained on-board monitoring data of a bulk carrier of ship length abt. 180m. for one year and a half between March 2017 and July 2018.

To analyze on-board monitoring data, physical model simulation can be applied in general. However, it is difficult to predict ship performance accurately in actual sea by physical model simulation. Therefore, the author proposed (habu, 2020) a way of analyzing on-board monitoring data using Gaussian process regression (GPR), which is one of machine learning and uses single-fidelity data. It is shown that GPR can directly predict ship speed (Vs) and shaft horsepower (SHP) from measured voyage data and shows good agreement with measured Vs and SHP. This estimator cannot show reasonable result of speed-power curves in comprehensively external condition according to Beaufort scales (BF) condition which the ship hasn't encountered. It means that the condition is extrapolation condition for the estimator. To solve those problems, the author applies Multi-fidelity Gaussian process regression (MF-GPR).

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