Comparison of supervised machine learning methods to predict ship propulsion power at sea. (1st February 2022)

• Record Type: Journal Article
• Authors: Lang, Xiao; Wu, Da; Mao, Wengang
• Is Part Of: Ocean engineering. Volume 245(2022)
• Publication Date: 2022-02-01
• Subjects: Supervised machine learning -- Ship propulsion power -- Metocean environments -- Full-scale measurements -- XGboost; Ocean engineering -- Periodicals; Ocean engineering; Periodicals; 620.4162
• DOI: 10.1016/j.oceaneng.2021.110387
• Languages: English
• ISSNs: 0029-8018

Abstract

Abstract: As the shipping moves towards digitization, a large amount of ship energy performance-related information collected during a ship's sailing provides opportunities to derive data-driven performance models using different machine learning algorithms. This paper compares several typical supervised machine learning algorithms, i.e., eXtreme Gradient Boosting (XGBoost), artificial neural network, support vector machine, and statistical regression methods, for the ship speed–power modeling. First, a general data pre-processing framework is presented. The different machine learning based models are trained by both ship operational parameters and encountered metocean conditions. Based on the full-scale measurement data collected at two types of worldwide sailing ships, the pros and cons of different machine learning algorithms for the ship's speed–power performance modeling are compared. Finally, the best performed XGboost model is chosen to analyze the sensitivity due to the amount of available ship data, assumed time period for each stationary waypoint (data sample) used for the model training, and their impact on online performance prediction. Highlights: Compare several supervised machine learning algorithms and statistical regression methods for ship speed–power modeling. A generic ship full-scale measurement pre-processing framework is proposed. Ship operational parameters and encountered metocean conditions are applied as training features. Sensitivity due to the data volume, stationary period of each data sample, and online performance prediction are analyzed.