표제지
목차
ABSTRACT 6
제1장 서론 16
1.1. 연구의 배경 16
1.2. 연구의 목적 27
1.3. 연구의 방법 및 구성 30
제2장 선행연구 분석 34
2.1. 의사결정지원 시스템 34
2.1.1. 의사결정의 과정 및 유형 35
2.1.2. 의사결정지원시스템의 구성 및 분류 39
2.1.3. 지능형 의사결정 지원 시스템 43
2.2. 선박교통관제시스템의 변화 49
2.2.1. 선박교통관제시스템 50
2.2.2. 지능형 해상교통정보시스템 56
2.2.3. 선박교통관제사의 의사결정 75
제3장 데이터마이닝과 선박충돌 위험도 평가 85
3.1. AIS 데이터 마이닝 85
3.1.1. AIS 데이터의 특징 86
3.1.2. AIS 데이터 마이닝 89
3.1.3. 해상교통 패턴 마이닝 91
3.2. 선박 충돌 위험도 평가 97
3.2.1. 선박 충돌 위험도 평가 개요 97
3.2.2. 미시적 선박 충돌 위험도 평가 100
3.2.3. 거시적 선박 충돌 위험도 평가 115
제4장 Pindex 식별 알고리즘 생성을 위한 전처리 방법 및 퍼지추론 시스템 120
4.1. AIS 데이터 전처리 123
4.1.1. AIS 데이터 Screening 123
4.1.2. AIS 데이터 cleaning 125
4.1.3. AIS 데이터 분류 126
4.1.4. 선박 항적 데이터 위치 보간 127
4.2. 선박항적 군집화 방법 및 개요 132
4.2.1. DBSCAN Algorithm의 정의 133
4.2.2. 클러스터 매개변수 설정 138
4.2.3. 군집 내 조우선박 충돌 위험도 평가 141
4.3. 퍼지 추론 기반 선박충돌위험 식별 144
4.3.1. 퍼지추론 시스템 개요 144
4.3.2. 퍼지기반 충돌위험지수 추론 시스템 153
제5장 Pindex 식별 알고리즘의 개발 및 실 해역 적용 160
5.1. 대상해역 선정 160
5.2. AIS Data-Preprocessing 162
5.2.1. AIS Data Screening 163
5.2.2. AIS data cleaning 166
5.2.3. 선박 항적 데이터 위치 보간 168
5.3. DBSCAN 알고리즘을 적용한 선박항적 군집화 172
5.3.1. 데이터 세트 선정 172
5.3.2. 매개변수 설정 174
5.4. T-K FIS를 사용한 Pindex 식별 183
5.4.1. Cluster 내 조우선박 간 충돌위험 계산 184
5.4.2. 입력변수의 퍼지화 187
5.4.3. IF-THEN 규칙 생성 190
5.4.4. T-K FIS 적용 Pindex 식별 192
5.4.5. 전체해역의 Pindex 식별 194
5.5. 알고리즘의 검증 200
5.5.1. 기준시간 변화에 따른 알고리즘 적용 결과 200
5.5.2. 전문가 집단에 의한 알고리즘 성능 검증 206
제6장 결론 212
6.1. 연구의 결론 212
6.2. 연구의 한계 및 향후 연구과제 218
부록 A_Cluster ID에 따른 Pindex 출력 결과 220
참고문헌 250
Table 1.1. Milestones for maritime safety 17
Table 1.2. Studies show that human error affects marine accidents 22
Table 1.3. Marine accidents by sea area over the past 5 years 29
Table 2.1. Intelligent decision support systems using AI 48
Table 2.2. List of proposed Maritime Services for use in MSP 58
Table 2.3. R&D items of core technology for Korean e-Nav. 65
Table 2.4. Strategy and implementation for e-Nav. 70
Table 2.5. Job competency unit of VTSO based on NCS 77
Table 3.1. Attributes of AIS data 87
Table 3.2. Class A type shipborne mobile equipment Reporting Intervals 88
Table 4.1. Types of fuzzy membership function 146
Table 4.2. Design methods for FIS 148
Table 4.3. The function F describing the degree of DNH 155
Table 4.4. Collision Avoidance by each level 156
Table 4.5. Reasoning rules of level of CRI 159
Table 5.1. Mokpo VTS area 161
Table 5.2. Results of AIS data pre-processing 166
Table 5.3. Time interpolation value 169
Table 5.4. Course interpolation value 170
Table 5.5. Number of noise point with varied є and MinPts 176
Table 5.6. Number of Cluster ID with varied є and MinPts=5 178
Table 5.7. Collision Avoidance by each level 183
Table 5.8. Number of input parameter 186
Table 5.9. Range of DCPA/TCPA by collision risk level 189
Table 5.10. fuzzy inference table for Pindex 190
Table 5.11. Components of the fuzzy inference rules 191
Table 5.12. Pindex of Cluster 1 193
Table 5.13. Sort all clusters in descending order of Pindex 195
Table 5.14. AIS data for 10 pairs of high-risk vessels 196
Table 5.15. Range of DCPA/TCPA by collision risk level 201
Table 5.16. Sort all clusters in descending order of Pindex 202
Table 5.17. Sort all clusters in descending order of Pindex 203
Table 5.18. Composition of a group of experts 206
Table 5.19. High-risk vessel identification results by groups 207
Table 5.20. The standard SUS 209
Table 5.21. SUS score range 210
Table 5.22. Results of applying a transform expression to the user response results 211
Fig. 1.1. Flowchart of the Study Process 31
Fig. 2.1. Description of Simon's model(1977, 1997) of the process of DSS 38
Fig. 2.2. DSS Components of Sage(2001) 41
Fig. 2.3. Structure of a decision support system 46
Fig. 2.4. Overarching e-Navigation Architecture 59
Fig. 2.5. Elemental description of e-Nav 60
Fig. 2.6. Smart-Navigation concept diagram 64
Fig. 2.7. Concept of core technology for Korean e-Nav 65
Fig. 2.8. Concept of intelligent maritime traffic information service 70
Fig. 2.9. DST increasing alerts 82
Fig. 3.1. The moving vector diagram of encounter ships 101
Fig. 3.2. Definition of geometrical collision diameter Dij[이미지참조] 104
Fig. 3.3. Crossing waterways with risk area of ship-ship collision 106
Fig. 3.4. Different domain-based safety criteria 108
Fig. 3.5. Illustration of room to maneuver 112
Fig. 3.6. Illustration of CTPA 113
Fig. 3.7. Visualization of near-miss density 116
Fig. 3.8. Locations of accidents 117
Fig. 3.9. Visualization of ship conflct Hot-Spots 118
Fig. 3.10. Visualization of ship clustering and TCR of random ship 119
Fig. 4.1. Pindex identification process 122
Fig. 4.2. Protocal for AIS data transmission 130
Fig. 4.3. Position interpolation by standard time interval 130
Fig. 4.4. Concept used in the DBSCAN 133
Fig. 4.5. Illustrates the concept of density-reachability and density-connectivity 136
Fig. 4.6. Pseudocode of DBSCAN Algorithm 137
Fig. 4.7. Density varied datapoints 139
Fig. 4.8. Comparison of risk perceptions of ship traffic situations in the identical conditions 141
Fig. 4.9. Identify the collision risk of multi-vessels 142
Fig. 4.10. Process of Fuzzy inference system 146
Fig. 4.11. Membership function of fuzzy parameter 147
Fig. 4.12. Fuzzy Reasoning of the T-K FIS with two-inputs 150
Fig. 5.1. Mokpo VTS area(KGC) 161
Fig. 5.2. Distribution of ship trajectories from Mokpo VTS 162
Fig. 5.3. AIS data extraction by Speed limit 164
Fig. 5.4. AIS data extraction by Speed limit 165
Fig. 5.5. AIS data extraction by cleaning 167
Fig. 5.6. Time Normalization of 86,400s to value between 0 and 1 168
Fig. 5.7. AIS data extraction by interpolation 171
Fig. 5.8. Distribution of ship's position by reference time 172
Fig. 5.9. Distribution of AIS data by reference time tk[이미지참조] 173
Fig. 5.10. Ship distribution over reference time 173
Fig. 5.11. Points sorted by distance to k-th nearest neighbor 175
Fig. 5.12. Relationships between є, MinPts and noise point 177
Fig. 5.13. Data distribution of Maximum data 178
Fig. 5.14. Results of DBSCAN with varied є and MinPts=5 179
Fig. 5.15. Result of DBSCAN with є=0.015 and MinPts=5 180
Fig. 5.16. ship distribution in each cluster ID(1) 181
Fig. 5.17. ship distribution in each cluster ID(2) 182
Fig. 5.18. Proposed FIS for designing the Pindex 183
Fig. 5.19. Encounters situation 184
Fig. 5.20. Modified membership functions; a)DCPA, b)TCPA 189
Fig. 5.21. T-K FIS model using Matlab fuzzy logic tool box 192
Fig. 5.22. Pindex extraction 197
Fig. 5.23. Pindex extraction in Cluster 2 198
Fig. 5.24. Pindex extraction in Cluster 5 &13 198
Fig. 5.25. High risk vessels in VTS screen example 199
Fig. 5.26. Distribution of AIS data according to the reference time 200
Fig. 5.27. Result of DBSCAN at t385 with є=0.015 and MinPts=3 201
Fig. 5.28. Pindex extraction at t385[이미지참조] 204
Fig. 5.29. High risk vessels in VTS screen example 205
Fig. 5.30. Time taken by group 207
Fig. 5.31. Box graph based on experience within a group 208
Fig. 5.32. Score average by question 211