표제지
국문요약
Abstract
목차
1. 서론 17
1.1. 연구 배경 17
1.2. 연구 해역 19
2. 자료 및 방법 22
2.1. 자료 22
2.2. 방법 23
2.2.1. 음선 추적 실험 24
2.2.2. 인공신경망 모델 27
2.2.3. EOF 적용 역계산 30
3. 결과 37
3.1. KOOS 모델 유속 검증 결과 37
3.2. 역계산 검증 결과 39
3.2.1. 경로를 따르는 방향의 유속 검증 39
3.2.2. 수직 및 수평 유속장 검증 44
4. 토의 53
4.1. 3차원 유속 RMSE 및 PVE 장 53
4.2. 모델 성능 비교 60
4.3. 인공신경망 모델 잡음 실험 결과 65
5. 결론 68
6. 부록 70
참고문헌 78
Table 1. Simulation paths (numbers) and its matching points on each side (west (Yeosu), east (Namhae)) 19
Table 2. Locations of CAT stations 20
Table 3. Summary of KOOS model data 22
Table 4. Layers of the KOOS model and averaged layers of inverse estimation 31
Table 5. Explained Variance of EOF Mode 1 ~ 5 and its sum 31
Table 6. Summary of ADCP observations 37
Table 7. RMSE (unit: cm/s) and R (correlation coefficient) between the true value (KOOS model outputs) and results from... 40
Table 8. Average of RMSE at each layer 53
Table 9. RMSE from Bellhop + Inv. method and AI + Inv. method in July and August, 2022 64
Table 10. Correlation coefficients (R) between the true velocity and estimated velocity for Bellhop + Inv. method and AI + Inv.... 64
Table 11. CAT observation and Bellhop Model output data used for determining standard deviation 66
Table 12. RMSE of the along-path velocity between true value (KOOS model output) and the results from AI model noise test 67
Appendix 2. Overview of the data used for AI model inputs 71
Appendix 4. Summary of AI model build and hyperparameter 73
Appendix 6. RMSE mean of the along-path velocity between true value (KOOS model) and the results from AI model 75
Fig. 1. Study area in the Yeosu Bay, Korea. Black dots in (c) indicate the CAT stations and the lines connected between the... 21
Fig. 2. Flowchart of estimating three-dimensional current field. The gray-shaded boxes indicate each procedure, and non-shaded... 23
Fig. 3. An example of Bellhop ray simulation between K4 and K3. Red and black curves respectively indicate the reciprocal ray propagation paths. 26
Fig. 4. Three-dimensional EOF analysis results. (a) Eigenvectors for the first 3 modes at (a) Layer 1, (b) Layers 2 and 3, and (c) Layers 4 and 5. Time series in (a) represent the principle... 32
Fig. 5. Comparison of along-path velocity between ADCP Observations and KOOS model outputs. Blue and red dots indicate ADCP observations carried out in 2022 and 2023, respectively. R... 38
Fig. 6. Comparison between the true value (KOOS model outputs, black) and results from the inverse estimation (blue) at the path between K1 and K3. 41
Fig. 7. The same to Fig. 6 except between K4 and K3. 42
Fig. 8. The same to Fig. 6 except between K4 and K5. 43
Fig. 9. Snapshots of V-component along the section between simulation points. The figures in the left column are true value... 45
Fig. 10. The same to Fig. 9 except at 15 h on August 20, 2022. 46
Fig. 11. The same to Fig. 9 except for U-component at 12 h on August 9, 2022. 47
Fig. 12. The same to Fig. 11 except at 6 h on August 16, 2022. 48
Fig. 13. (Upper panels) Snapshots of mapped current velocity at layers 1, 3, and 4 at 12 h on July 15, 2022. Magenta, and blue arrows indicate the results from the inverse estimation and the... 49
Fig. 14. The same to Fig. 13 except at 15 h on July 11, 2022. 50
Fig. 15. The same to Fig. 13 except at 15 h on August 20, 2022. 51
Fig. 16. The same to Fig. 13 except at 18 h on August 23, 2022. 52
Fig. 17. RMSE of U-component velocity in July and August 2022. Black lines indicate simulation paths, and gray-shaded parts are land. 54
Fig. 18. The same to Fig. 17 except for V-component. 55
Fig. 19. The same to Fig. 17 except in January and February 2022. 56
Fig. 20. The same to Fig. 19 except for V-component. 57
Fig. 21. PVE of U-component and V-component for Layer 1, 3, and 4 in July and August 2022. Black lines indicate simulation paths, and gray-shaded parts are land. 59
Fig. 22. Comparison of the true value, results from 'Bellhop + Inv.' method, and results from 'AI + Inv.' method. The scatter plot shows July and August, and time series shows August. 61
Fig. 23. The same to Fig. 22 except for K4-K3 path. 62
Fig. 24. The same to Fig. 22 except for K4-K5 path. 63
Appendix 1. Schematic diagram of Bellhop model. White and red trapezoids mean Bellhop input and output, respectively. Blue... 70
Appendix 3. Map of information for AI model training data. Dotted line is the domain of numerical model which includes the tidal... 72
Appendix 5. Schematic diagram for AI model structure. White and red trapezoids mean input and target, respectively. Orange boxes... 74
Appendix 7. RMSE contour of the along-path velocity between true value (KOOS model) and the results from AI model 76
Appendix 8. Bias contour of the along-path velocity between true value (KOOS model) and the results from AI model 77