Data Analysis and Modeling of Ship Motion Data for Offshore Operations

Abstract

Ship intelligence aims to make the marine and offshore industries more efficient, innovative, and adaptable to future operations. In fact, ship intelligence has been listed as an important part of the digital agenda, one of the pillars of the European growth strategy. This is in part because of increasing interest in development and employment of autonomous ships. Autonomous ships use intelligence to make decisions that increase the control precision, lower the fuel consumption, and extend the operational window. Autonomous ships face greater challenges than autonomous cars, mainly because of the more complicated environment at sea. Wind and waves are the most vexing aspects of this environment. This calls for novel algorithms for the data analysis and modeling of ship motion and the environment. The present thesis examines data-driven analysis and modeling. Specif_- cally, it presents novel algorithms for surrogate model-based sensitivity and uncertainty analysis for data-driven ship motion modeling and sea state estimation driven by ship motion data. The research for this thesis is based on ship motion data. The sensitivity and uncertainty analysis relies on the sensor data from numerous offshore applications as input and the designated metric as output, which makes it possible to quantify how much the input contributes to the output. The result of sensitivity and uncertainty analysis can benefit for further understanding of the ship motion data and constructing compact ship motion models. As for sea state estimation driven by the ship motion data, deep learning based methods would be utilized that mine the intrinsic features of the motion data. To achieve the research goal of this thesis, a framework of sensitivity and uncertainty analysis for ship motion modeling and deep learning based sea state estimation models is proposed. Based on the proposed framework of sensitivity and uncertainty analysis for ship motion modeling, three sensitivity analysis methods and two uncertainty analysis methods are utilized and compared. To investigate the feasibility of sea state estimation based on ship motion data, two deep learning based models for sea state estimation have been introduced with the aim of identifying wave height and wave direction. To further extend the generality of deep learning models, a transfer learning based model for sea state estimation is proposed. The experimental results show the proposed framework of sensitivity and uncertainty analysis works well for identifying the important parameters of data sets with and without environmental factors. The experimental results of sensitivity analysis (SA)-based applications imply that the SAbased methods can be an useful tool in modeling offshore operations. The deep learning model based on dynamic positioning (DP) motion data verifies the feasibility of estimating the wave height. The deep learning model based on zigzag motion data illustrates the feasibility of identifying wave height and wave direction simultaneously. The experimental results with a transfer learning based model show it can address the weaknesses of the deep learning based sea state estimation model, which still lacks of generality