Application of Machine and Deep Learning to Mooring, Dynamic Positioning, and Ship Berthing Systems
DC Field | Value | Language |
---|---|---|
dc.contributor.advisor | 이승재 | - |
dc.contributor.author | 이대수 | - |
dc.date.accessioned | 2021-01-31T08:39:56Z | - |
dc.date.available | 2021-01-31T08:39:56Z | - |
dc.date.issued | 2020 | - |
dc.identifier.uri | http://repository.kmou.ac.kr/handle/2014.oak/12474 | - |
dc.identifier.uri | http://kmou.dcollection.net/common/orgView/200000342184 | - |
dc.description.abstract | In recent years, there have been a surge of advances in machine and deep learning due to accessibility to a large amount of digital data, developments in computer hardware, and state-of-the-art machine and deep learning algorithms proposed. The robust performance of the recent machine and deep learning algorithms have been proven in many applications such as natural language processing, computer vision, market research, self-driving car, autonomous shipping, and so on. The application of machine and deep learning is very powerful in a sense that one does not need to build such a complex and hard-coded system to implement sophisticated functionality. Instead, a machine and deep learning-based system can be trained on a collected training dataset and the trained system can robustly perform as desired. There are two main advantages of the use of machine and deep learning-based systems over the traditional hard-coded systems. First, as mentioned, the machine and deep learning-based systems do not require such complex and hard-coded algorithms, therefore, such learning systems are less prone to errors and faster to implement without much debugging. Second, the machine and deep learning-based systems can adapt to varying circumstances through re-training based on collected data. An example of the varying circumstance can be a varying purchase trend impacted by the media. Therefore, even if the input distribution from the circumstance changes over time, the machine and deep learning-based systems can easily adapt. In this paper, the machine and deep learning algorithms are applied to various applications such as a mooring system, dynamic positioning system (DPS), and ship berthing system. Specifically, the machine and deep learning algorithms are utilized to build a mooring line tension prediction system, a feed-forward system for DPS, an adaptive proportional-integral-derivative (PID) controller for DPS, and an automatic ship berthing system. | - |
dc.description.tableofcontents | 1. Introduction 1 2. Background of Machine and Deep Learning 4 2.1 Machine Learning 4 2.2 Deep Learning 9 2.2.1 Types of Deep Learning Layers 9 2.2.2 Activation Function and Weight Initialization Methods 18 2.2.3 Optimizers 19 2.2.4 Training Dataset Scaling 26 2.2.5 Transfer Learning 28 2.3 Reinforcement Learning 28 3. Machine Learning-Based Mooring Line Tension Prediction System 39 3.1 Introduction 39 3.2 Brief Comparison Between Conventional and Proposed Mooring Line Tension Prediction Systems 40 3.3 Proposed K-Means-Based Sea State Selection Method 41 3.3.1 Padding 42 3.3.2 K-Means 44 3.3.3 K-Means-Based Monte Carlo Method 45 3.3.4 Feature Vector Generation 47 3.3.5 Clustering of Relevant Sampled Sea States with K-Means 48 3.4 Proposed Hybrid Neural Network Architecture 50 3.4.1 Architecture 50 3.4.2 Training Procedure 54 3.5 Simulation and Result Discussion 55 3.5.1 Simulation Conditions 55 3.5.2 Overall Hs-focused NN model 56 3.5.3 Effectiveness of Batch Normalization 59 3.5.4 Low Hs-focused NN model 60 3.5.5 Proposed Hybrid Neural Network Architecture 61 4. Motion Predictive Control for DPS Using Predicted Drifted Ship Position Based on Deep Learning and Replay Buffer 65 4.1 Introduction 65 4.2 PID Feed-Back System and Wind Feed-Forward System 66 4.3 Proposed Motion Predictive Control 69 4.4 Numerical Modeling of Target Ship's Behavior 73 4.4.1 Target Ship and DPS 73 4.4.2 Equation of Motion of Target Ship 74 4.5 Effectiveness of Proposed Algorithms 76 4.5.1 Simulation Conditions 76 4.5.2 Types of Deep Learning Layers 77 4.5.3 Real-Time Normalization Method 78 4.5.4 Replay Buffer 80 4.6 Simulation and Result Discussion 81 4.6.1 Simulation Under One Environmental Condition 81 4.6.2 Simulation Under Two Different Sequential Environmental Conditions 84 5. Reinforcement Learning-Based Adaptive PID Controller for DPS 88 5.1 Introduction 88 5.2 Target Ship and DPS 90 5.2.1 PID Control in DPS 91 5.2.2 Hydrodynamics Associated with a Drifting Motion of a Ship 93 5.3 Proposed Adaptive Fine-Tuning System for PID Gains in DPS 95 5.4 Simulation Results 99 5.4.1 Effectiveness of the Proposed Adaptive Fine-Tuning System 99 5.4.2 Overall Performance Assessment 103 5.5 Discussion 107 6. Application of Recent Developments in Deep Learning To ANN-based Automatic Berthing System 111 6.1 Introduction 111 6.2 Mathematical Model of Ship Maneuvering 112 6.2.1 Mathematical Model for Ship-Maneuvering Problem 113 6.2.2 Modeling of Propeller and Rudder 114 6.3 Artificial Neural Network and Important Factors in Training the Network 115 6.3.1 Artificial Neural Network 115 6.3.2 Optimizer 117 6.3.3 Input Data Scaling 117 6.3.4 Number of Hidden Layers 118 6.3.5 Overfitting Prevention 118 6.4 Application of Recent Developments in Deep Learning to Automatic Berthing 119 6.5 Simulation and Result Discussion 125 7. Conclusion 131 7.1 Machine Learning-Based Mooring Line Tension Prediction System 131 7.2 Motion Predictive Control for DPS Using Predicted Drifted Ship Position Based on Deep Learning and Replay Buffer 132 7.3 Reinforcement Learning-Based Adaptive PID Controller for DPS 133 7.4 Application of Recent Developments in Deep Learning to ANN-Based Automatic Berthing System 134 | - |
dc.language | eng | - |
dc.publisher | 한국해양대학교 대학원 | - |
dc.rights | 한국해양대학교 논문은 저작권에 의해 보호받습니다. | - |
dc.title | Application of Machine and Deep Learning to Mooring, Dynamic Positioning, and Ship Berthing Systems | - |
dc.type | Dissertation | - |
dc.date.awarded | 2020. 8 | - |
dc.contributor.department | 대학원 조선해양시스템공학과 | - |
dc.contributor.affiliation | 한국해양대학교 대학원 조선해양시스템공학과 | - |
dc.description.degree | Master | - |
dc.identifier.bibliographicCitation | 이대수. (2020). Application of Machine and Deep Learning to Mooring, Dynamic Positioning, and Ship Berthing Systems | - |
dc.subject.keyword | Machine learning | - |
dc.subject.keyword | Deep learning | - |
dc.subject.keyword | Mooring system | - |
dc.subject.keyword | Dynamic positioning system (DPS) | - |
dc.subject.keyword | Automatic berthing system | - |
dc.identifier.holdings | 000000001979▲200000001758▲200000342184▲ | - |
Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.