설명가능한 인공지능기반 기상예측 인과관계 도출 기술 개발 = Development of explainable AI-based methods for causal relation estimation in AI-based weather forecast models. 3 / 국립기상과학원, 인공지능기상연구과 [편]

국회도서관 홈으로 정보검색 소장정보 검색

결과 내 검색

동의어 포함

고급검색

상세검색
저자 검색
관련 키워드 검색
주제별 검색

완전일치
전방일치
후방일치

인명/단체명

	저자정보	상세정보
인명/단체명을 입력하세요.

전방일치
완전일치
후방일치
부분일치

키워드

대표어
외국어
네이버 백과사전

용어관계 검색결과
대표어	동의어	상위어	하위어	관련어	대립어

대분류

중분류

소분류

소장자료
외부기관 자료

일반도서 설명가능한 인공지능기반 기상예측 인과관계 도출 기술 개발 = Development of explainable AI-based methods for causal relation estimation in AI-based weather forecast models. 3

저자사항
국립기상과학원, 인공지능기상연구과 [편]
발행사항
서귀포 : 국립기상과학원 인공지능기상연구과, 2024
청구기호
551.630285 -25-1
자료실 서울관 인문자연과학자료실(314호) 도서위치안내[서울관]
형태사항
xxv, 196 p. : 삽화, 도표, 서식 ; 30 cm
총서사항
기상업무지원기술개발연구 = Research and development for KMA weather, climate, and earth system services
표준번호/부호
ISBN: 9791193905241
제어번호
MONO12025000012342
주기사항
본문은 한국어, 요약문과 목차는 한국어, 영어가 혼합 수록됨
연계정보
원문
외부기관 원문

목차보기

[표지]

연구 보고서

요약문 25

Summary 28

제1장 개요 31

제1절 연구개발의 목표 및 필요성 31

1. 연구개발 목표 31

2. 연구개발 필요성 31

제2절 연구개발 수행범위 33

제2장 초단기 강수예측 모델 설명가능 인공지능 알고리즘 개발 35

제1절 XAI 알고리즘 적용을 위한 AI-초단기 강수예측 모델 특성 분석 35

1. 연구개발 목표 35

2. 1차년도 연구 내용 35

제2절 다중 기상정보 초단기 강수예측 모델의 인과관계 설명 기법 개발 38

1. 연구개발 목표 38

2. 2차 년도 연구 내용 38

3. 3차년도 연구 내용 49

제3절 소결 102

제3장 예보관 의사결정 지원 설명 기법 개발 103

제1절 XAI 구조설계를 위한 유저데이터 선정·수집·분석 103

1. 연구개발 목표 103

2. 1차년도 연구 내용 103

제2절 예보관 의사결정 메커니즘 기반 초단기 강수예측 해석모델 개발 108

1. 연구개발 목표 108

2. 1차년도 연구 내용 108

3. 2차년도 연구 내용 116

4. 3차년도 연구 내용 131

제3절 예보관 피드백 적용 기술 개발 151

1. 연구개발 목표 151

2. 2차년도 연구 내용 151

제4절 예보관 의사결정 지원을 위한 최적 XAI 모형 및 HCI 구조 설계 154

1. 연구개발 목표 154

2. 1차년도 연구 내용 154

3. 2차년도 연구 내용 157

4. 3차년도 연구 내용 162

제5절 소결 164

제4장 결론 165

참고문헌 166

부록 178

A. 퇴직 예정 예보관 대상 설명 수요 조사를 위한 구조화된 설문지 178

B. 설명 인터페이스 시스템 사용자 경험 조사 설문지 186

C. 사례 기반 컨셉 설명 인터페이스 시스템 사용자 경험 조사 설문지 195

D. 컨셉 추출 재료 비교 : 활성화 값과 기여도 값 201

E. 비지도식 컨셉 추출 결과 사례(2차년도) 202

F. 비지도식 컨셉 추출 결과 사례(3차년도) 209

G. 예보분석서 기반 강수 메커니즘 컨셉 데이터셋 예시 212

H. 세그멘테이션 모델을 위한 Wrapper 함수 224

[뒷표지] 227

List of Tables

Table 1.1. Research scope in the three years' of requests for proposals 33

Table 1.2. Research strategies and detail research scope 34

Table 2.1. Prediction results using only radar input 43

Table 2.2. Prediction results using radar, BTD1, and BTD4 channel 43

Table 2.3. Definition of LRP rule for each layer used in the experiment 44

Table 2.4. Description of the variables of ERA5 49

Table 2.5. Definition of BTD Numbers 50

Table 2.6. Summary of modeling experiments 52

Table 2.7. Performance scores of modeling experiments.(red and blue represents highest... 53

Table 2.8. The date of cases for each rainfall type(KST) 57

Table 2.9. Key input features for each class in the low-pressure rainfall case... 59

Table 2.10. Key input features for each class in the low-pressure rainfall case... 61

Table 2.11. Key input features for each class in the convective... 70

Table 2.12. Key input features for each class in the convective rainfall case... 71

Table 2.13. Key input features for each class in the orographic rainfall case... 82

Table 2.14. Key input features for each class in the frontal rainfall case(2020.09.17.... 92

Table 2.15. Key input features for each class in the frontal rainfall case(2020.06.24.... 94

Table 3.1. Definition of desired explanation for forecasting and selectiong of explainable AI methods 104

Table 3.2. Expected calibration error(ECE) of calibrated confidence on... 112

Table 3.3. Pseudo labels which appears frequently in each cluster 120

Table 3.4. P-value of pairwise t-Test on each concept 126

Table 3.5. Accuracy of query algorithms comparing on four types of distance metrics 129

Table 3.6. Accuracy of query algorithms 129

Table 3.7. References related to the rainfall mechanism classification in Korea 144

Table 3.8. Types of rainfall mechanism concept dataset labels 144

Table 3.9. Runtime comparison of principal neuron component-based neighbor search engine(PC-NSE) 147

Table 3.10. The Importance scores on the number of samples for each concept class. The... 148

List of Figures

Fig. 1.1. Damage costs by type of natural disaster worldwide. 31

Fig. 1.2. The need for explainable artificial intelligence in weather forecasting environment.... 32

Fig. 1.3. The need for explainable AI weather forecasting 33

Fig. 1.4. Three years of research process 34

Fig. 2.1. Convolutional neural networks-based rainfall forecasting model structure 35

Fig. 2.2. An example of input data and labels 36

Fig. 2.3. Input domain for the study. The... 37

Fig. 2.4. The range of input values of train dataset 37

Fig. 2.5. Differences in spatial distribution of input contribution according to... 42

Fig. 2.6. Jangma and typhoon cases. Comparison of observed(GT) and forecast... 45

Fig. 2.7. Confusion matrix for Jangma and Typhoon cases 46

Fig. 2.8. Comparison of true values of radar input for each input time period in the... 47

Fig. 2.9. Comparison of true values of satellite input(IR105) for each input time... 47

Fig. 2.10. Comparison of true values of satellite input(BTD4) for each input time... 48

Fig. 2.11. Comparison of true values of satellite input(BTD1) for each input time... 48

Fig. 2.12. Visualization of ERA5 Single level(KST 2020.09.06. 09:00) 50

Fig. 2.13. Visualization of ERA5 Pressure level(KST 2020.09.06. 09:00) 50

Fig. 2.14. Visualization of the original data of GK2A IR105, IR112, WW0B3 channels(KST... 51

Fig. 2.15. BTD channel visualization(KST 2020.09.06. 09:00) 51

Fig. 2.16. Comparison of regions between GK2A image(Gray) and ERA5(Color) 53

Fig. 2.17. Cyclonic rainfall KST 2020.06.29. 21:00 54

Fig. 2.18. Convective type KST 2020.07.29. 17:00 55

Fig. 2.19. Orographical effect type KST 2020.08.11. 09:00 55

Fig. 2.20. Frontal type KST 2020.09.17. 15:00 56

Fig. 2.21. The two cases for analyzing the feature contributions in the low-pressure rainfall 58

Fig. 2.22. Distribution of the top 5 feature contributions over time for each class... 59

Fig. 2.23. Distribution of the top 5 feature contributions over time for each class... 60

Fig. 2.24. Comparison of key feature distributions over time in two... 61

Fig. 2.25. Spatial distribution of major feature contributions for class1 in the... 63

Fig. 2.26. Spatial distribution of major feature contributions for class2 in the... 64

Fig. 2.27. Common and individual spatial distributions of key features over time for... 65

Fig. 2.28. Common and individual spatial distributions of key features over time for... 65

Fig. 2.29. Spatial distribution of major feature contributions for class1 in the... 66

Fig. 2.30. Spatial distribution of major feature contributions for class2 in the... 67

Fig. 2.31. Common and individual spatial distributions of key features over time for... 68

Fig. 2.32. Cyclonic rainfall cases(2020.07.10. 06:00 KST). The spatial... 68

Fig. 2.33. The two cases for analyzing the feature contributions in the convective rainfall 69

Fig. 2.34. Distribution of the top5 feature contributions over time for each class in the... 70

Fig. 2.35. Distribution of the top 5 feature contributions over time for each class in the... 71

Fig. 2.36. Comparison of key feature distributions over time in two convective... 72

Fig. 2.37. Spatial distribution of major feature contributions for class1 in the... 73

Fig. 2.38. Spatial distribution of major feature contributions for class2 in the... 74

Fig. 2.39. Common and individual spatial distributions of key features over time for... 75

Fig. 2.40. Common and individual spatial distributions of key features over time for... 75

Fig. 2.41. Spatial distribution of major feature contributions for class1 in the... 76

Fig. 2.42. Spatial distribution of major feature contributions for class2 in the convective... 77

Fig. 2.43. Common and individual spatial distributions of key features over time for... 78

Fig. 2.44. Common and individual spatial distributions of key features over time for... 78

Fig. 2.45. The two cases for analyzing the feature contributions in the orographic rainfall 79

Fig. 2.46. Distribution of the top 5 feature contributions over time for each class... 80

Fig. 2.47. Key input features for each class in the orographic rainfall case... 81

Fig. 2.48. Distribution of the top 5 feature contributions over time for each class in... 82

Fig. 2.49. Comparison of key feature distributions over time in two orographic rainfall cases 83

Fig. 2.50. Spatial distribution of major feature contributions for class1 in the... 84

Fig. 2.51. Spatial distribution of major feature contributions for class2 in the... 85

Fig. 2.52. Common and individual spatial distributions of key features over time for... 86

Fig. 2.53. Common and individual spatial distributions of key features over time for... 86

Fig. 2.54. Spatial distribution of major feature contributions for class1 in the orographic... 88

Fig. 2.55. Spatial distribution of major feature contributions for class2 in the orographic... 89

Fig. 2.56. Common and individual spatial distributions of key features over time for... 90

Fig. 2.57. Common and individual spatial distributions of key features over time for... 90

Fig. 2.58. The two cases for analyzing the feature contributions in the frontal rainfall 91

Fig. 2.59. Distribution of the top 5 feature contributions over time for each class in... 92

Fig. 2.60. Distribution of the top 5 feature contributions over time for each class in... 94

Fig. 2.61. Comparison of key feature distributions over time in two frontal rainfall cases 95

Fig. 2.62. Spatial distribution of major feature contributions for class1 in the frontal rainfall... 96

Fig. 2.63. Spatial distribution of major feature contributions for class2 in the frontal... 97

Fig. 2.64. Common and individual spatial distributions of key features over time for... 98

Fig. 2.65. Common and individual spatial distributions of key features over time for... 98

Fig. 2.66. Spatial distribution of major feature contributions for class1 in the frontal... 99

Fig. 2.67. Spatial distribution of major feature contributions for class2 in the frontal... 100

Fig. 2.68. Common and individual spatial distributions of key features over time for... 101

Fig. 2.69. Common and individual spatial distributions of key features over time for... 101

Fig. 3.1. Research process of development of user-centric explainable AI 104

Fig. 3.2. NCEP/DOE Reanalysis 2 model 2m tem perature 106

Fig. 3.3. NCEP/DOE Reanalysis 2 model K Index 106

Fig. 3.4. KIM GDAPS 850nhPa wind flow streamline and wind speed 106

Fig. 3.5. KIM GDAPS auxiliary weather chart and 850hPa convergence 106

Fig. 3.6. Examples of CAPPI and RHI 107

Fig. 3.7. Left : 1.5km height CAPPI radar reflection(shaded color) and horizontal wind distribution... 107

Fig. 3.8. GK2A Infrared Channel 105(Clear-Sky Window) 108

Fig. 3.9. GK2A Daytime Natural Color RGB Channel 108

Fig. 3.10. The structure of the precipitation classifier (a) and the resulting confusion matrix (b). The... 109

Fig. 3.11. Performance diagram. The diagram helps visualize the overall performance of bias, CSI,... 109

Fig. 3.12. The case of 2020-08-07 at 14:00 UTC with temperature scaling 112

Fig. 3.13. Feature attribution. The heatmap describes the location and the degree of relevance of... 113

Fig. 3.14. Quantitative evaluation of the output reasoning explan... 113

Fig. 3.15. Anecdotal evidence based on domain expert's case analysis of monsoon front and extratr... 114

Fig. 3.16. Survey results of the user experience of the prototype for an explainable artificial intellige... 115

Fig. 3.17. Survey results on the demand of 23 items of explainable artificial intelligence among... 116

Fig. 3.18. The need of instance segmentation for each rain... 118

Fig. 3.19. Manifold learning applied to the input feature space, with colors overlaid to... 120

Fig. 3.20. An example of post-forecast document for weather concept evaluation... 121

Fig. 3.21. Concept activation vector(υlc)) and its linear classifier[이미지참조] 122

Fig. 3.22. Histogram of the coefficients of linear classifiers using... 123

Fig. 3.23. Accuracy of each linear classifier on the test dataset. 124

Fig. 3.24. P-value of pairwise t-Test for each concept pair.... 126

Fig. 3.25. Frequency of the pixels of each target class of ground truth... 126

Fig. 3.26. Survey results of Q1 and Q2 of identi... 131

Fig. 3.27. Survey results of Q4 of comparing the... 131

Fig. 3.28. The heatmap of cosine similarities of human annotated concept label pairs. The cosin... 132

Fig. 3.29. The outline of multi-label deep clustering with soft-shilhouette coefficient 135

Fig. 3.30. Manifold of the embedding vector spaces at each stage of the proposed representation... 136

Fig. 3.31. Manifold of the embedding vector spaces colorized by three human-annotated label data... 137

Fig. 3.32. Examples of top-three nearest neighbors from four query samples measured by distance... 138

Fig. 3.33. Examples in the fine-tuned feature space. Four labels are visualized in this figure to analy... 139

Fig. 3.34. Examples of lake-effect-snowfall, sea-breeze-convergence, and east-coast-rainfall in th... 140

Fig. 3.35. Comparison of polar low and typhoon cases. The score refers to the probabilistic score... 141

Fig. 3.36. Additional identification test of concept prober #19 : a comparison between polar low cas... 141

Fig. 3.37. Sample questions form the structured questionnaires for user evaluation of extracted conc... 142

Fig. 3.38. Accuracy of classifying heterogeneous an... 143

Fig. 3.39. The number of samples of human-annotated concept labels. In the legend, 'POSTHOC'... 145

Fig. 3.40. Examples of neighbor search engine and probabilistic concept explanations. The firs... 148

Fig. 3.41. Importance score of concept activation vectors for each concept. The... 150

Fig. 3.42. Importance score of concept activation vectors(cont.) 150

Fig. 3.43. Perturbation test on the Ost concept activation vector 152

Fig. 3.44. Perturbation test on the 1st concept activation vector 152

Fig. 3.45. Perturbation test on the 2nd concept activation vector 152

Fig. 3.46. Perturbation test on the 4th concept activation vector 153

Fig. 3.47. Perturbation test on the 8th concept activation vector 153

Fig. 3.48. Perturbation test on the 11th concept activation vector 153

Fig. 3.49. Monitoring panel of Amazon SageMaker Clarify service's input data drift 154

Fig. 3.50. Monitoring panel of Amazon SageMaker Clarify service's input attribution change 155

Fig. 3.51. Monitoring panel of Google Cloud Console service's input attribution 155

Fig. 3.52. Human computer interaction interface of similar image search explanation system in medic 156

Fig. 3.53. Design plaan of visualization of explanation modules of model accuracy and... 157

Fig. 3.54. Design plan of visualization of input attribution module 157

Fig. 3.55. Use case diagram of user interface system 158

Fig. 3.56. Design and operational scenarios for the forecaster's decision support interface system architecture 159

Fig. 3.57. User interface prototype design proposal(1st alt.) 160

Fig. 3.58. User interface prototype design proposal(2nd alt.) 161

Fig. 3.59. Survey results of Q5 about the user experience of... 162

Fig. 3.60. Use case design for user interface 163

Fig. 3.61. User interface prototype design proposal(3rd alt.) 163

Fig. A.1. Structured questionnaire for survey focused on retiree-to-be forecasters 179

Fig. A.2. Structured questionnaire for survey focused on retiree-to-be forecasters(cont.) 180

Fig. A.3. Structured questionnaire for survey focused on retiree-to-be forecasters(cont.) 181

Fig. A.4. Structured questionnaire for survey focused on retiree-to-be forecasters(cont.) 182

Fig. A.5. Structured questionnaire for survey focused on retiree~to-be forecasters(cont.) 183

Fig. A.6. Structured questionnaire for survey focused on retiree-to-be forecasters(cont.) 184

Fig. A.7. Structured questionnaire for survey focused on retiree-to-be forecasters(cont.) 185

Fig. B.1. Questionnaire design for user experience on explainable AI interface system 186

Fig. B.2. Structured questionnaire for user experience on explainable AI... 187

Fig. B.3. Structured questionnaire for user... 188

Fig. B.4. Structured questionnaire for user experience on explainable AI... 189

Fig. B.5. Structured questionnaire for user experience on explainable AI... 190

Fig. B.6. Structured questionnaire for user experience on explainable AI... 191

Fig. B.7. Structured questionnaire for user experience on explainable AI... 192

Fig. B.8. Structured questionnaire for user experience on explainable AI... 193

Fig. B.9. Structured questionnaire for user experience on explainable AI... 194

Fig. C.1. Structured questionnaire for survey of user experience on example-based... 195

Fig. C.2. Structured questionnaire for survey of user experience on example-based... 196

Fig. C.3. Structured questionnaire for survey of user... 197

Fig. C.4. Structured questionnaire for survey of user... 198

Fig. C.5. Structured questionnaire for survey of user experience... 199

Fig. C.6. Structured questionnaire for survey of user experience... 200

Fig. E.1. Manifold topology of activation and relevance space. From left to right : result on... 201

Fig. E.2. Visualization of input feature data for each cluster(from top to bottom, corresponding... 202

Fig. E.3. Visualization of input feature data for each cluster(from top to bottom, corresponding... 203

Fig. E.4. Visualization of input feature data for each cluster(from top to bottom, corresponding... 204

Fig. E.5. Visualization of input feature data for each cluster(from top to bottom, corresponding... 205

Fig. E.6. Visualization of input feature data for each cluster(from top to bottom, corresponding... 206

Fig. E.7. Visualization of input data for each cluster(from top-left to bottom-right, corresponding... 207

Fig. E.8. Visualization of input data for each cluster(from top-left to bottom-right, corresponding... 208

Fig. F.1. The statistics of clustered instances on the human annotated concept evaluation labels. N... 209

Fig. F.2. Example instances of 24 clusters. The date time is in UTC. 210

Fig. F.3. Example instances of 24 clusters. The date time is in UTC.(Cont.) 211

Fig. F.4. The samples of changma concept in the weather evaluation label dataset. Changma is als... 212

Fig. F.5. The samples of stationary-front concept in the weather evaluation label dataset. A stationa... 213

Fig. F.6. The samples of cold-front concept in the weather evaluation label dataset. A cold front is... 214

Fig. F.7. The samples of warm-front concept in the weather evaluation label dataset. A warm front... 214

Fig. F.8. The samples of typhoon concept in the weather evaluation label dataset. Typhoons are int... 215

Fig. F.9. The samples of cyclonic concept in the weather evaluation label dataset. Cyclonic precipit... 216

Fig. F.10. The samples of scattered-shower concept in the weather evaluation label dataset. Scatte... 216

Fig. F.11. The samples of drizzle concept in the weather evaluation label dataset. Drizzle refers to... 217

Fig. F.12. The samples of convectional concept in the weather evaluation label dataset. Convection... 218

Fig. F.13. The samples of scattered-shower concept in the weather evaluation label dataset. Scatte... 219

Fig. F.14. The samples of sea-breeze-convergence concept in the weather evaluation label datase.... 219

Fig. F.15. The samples of east-coast-rainfall concept in the weather evaluation label dataset. East... 220

Fig. F.16. The samples of lake-effect-snowfall concept in the weather evaluation label dataset. Lak... 221

Fig. F.17. The samples of upper-low-jet-coupling concept in the weather evaluation label dataset.... 222

Fig. F.18. The samples of low-level-jet concept in the weather evaluation label dataset. Low-level... 223

Fig. F.19. The samples of maintenance concept in the weather evaluation label dataset. The continu... 224

AI 100자 요약 서비스

AI 100자 요약 서비스

인공지능이 요약한 요약문입니다. 본문과 일치하지 않은 내용이 있을 수 있으니, 원문을 확인하시기 바랍니다.

이 연구 보고서는 국립기상과학원의 인공지능기상연구과에서 진행한 설명 가능한 인공지능(XAI)을 활용한 기상 예측 기술 개발에 관한 최종 보고서입니다. 연구는 2022년부터 2024년까지 진행되며, 기상 예보의 신뢰성과 정확성을 높이기 위해 AI 모델의 설명 가능성을 강화하는 것을 목표로 합니다. 연구는 크게 세 가지 목표를 설정하였습니다. 첫째, 초단기 강수예측 모델의 구조와 데이터를 분석하여 최적의 XAI 모형을 선정하고, 예보관의 의사결정 메커니즘을 반영한 사용자 중심의 인터페이스를 설계합니다. 둘째, 예보관의 피드백을 반영하여 예측 결과 해석 기법을 개발하고, 예보관의 사용자 경험을 고려한 설명 결과 표출 방식을 설계합니다. 셋째, 예보관의 의사결정 지원을 위한 최적의 XAI 모형과 HCI 구조를 설계하여 기상청 시스템과 연계 가능한 인터페이스를 제안합니다. 부록에서는 퇴직 예정 예보관을 대상으로 한 설문조사와 사용자 경험 평가를 통해 AI 모델의 신뢰성과 설명 가능성을 높이기 위한 다양한 방법론을 제시합니다. 이 연구는 기상 예보 분야에서 AI 기술의 실무적 활용 가능성을 높이고, 재난 예방 및 중요한 기상 관련 의사결정 과정에서 폭넓게 활용될 수 있는 기반을 마련하는 데 기여할 것으로 기대됩니다.

※ 출처: 상단의 서지정보에서 확인
※ 원저작물에 적용된 공공누리 유형: 제2유형

이용현황보기

설명가능한 인공지능기반 기상예측 인과관계 도출 기술 개발 = Development of explainable AI-based methods for causal relation estimation in AI-based weather forecast models. 3 이용현황 표 - 등록번호, 청구기호, 권별정보, 자료실, 이용여부로 구성 되어있습니다.
등록번호	청구기호	권별정보	자료실	이용여부
0003178383	551.630285 -25-1	v.3	서울관 인문자연과학자료실(314호)	이용가능
0003178384	551.630285 -25-1	v.3	서울관 인문자연과학자료실(314호)	이용가능

자료명
저자사항
제어번호
*요청자 이름
*전화번호	휴대폰 번호를 입력하세요.
*이메일	@
*요청내용
*오류항목

* 서재명
설명
* 공개수준	비공개 완전공개 * 주의: 국회도서관 이용자 모두에게 공유서재로 서비스 됩니다.

고급검색

다국어입력

일반도서 설명가능한 인공지능기반 기상예측 인과관계 도출 기술 개발 = Development of explainable AI-based methods for causal relation estimation in AI-based weather forecast models. 3

목차보기

AI 100자 요약 서비스

이용현황보기

추천서가 (다양한 추천 자료를 만나보세요)

권호

알림톡 발송로 자료명, 기사명/저자명, 수록지명, 자료실, 서가번호, 전화번호로 구성되어 있습니다.




전화번호

연속간행물 상세정보 입니다.
청구기호
자료명/저자사항
발행사항
형태사항
ISSN

고급검색

다국어입력

일반도서 설명가능한 인공지능기반 기상예측 인과관계 도출 기술 개발 = Development of explainable AI-based methods for causal relation estimation in AI-based weather forecast models. 3

목차보기

AI 100자 요약 서비스

이용현황보기

추천서가 (다양한 추천 자료를 만나보세요)

MARC 보기

오류 데이터 정정요청

알림톡 발송

권호기사보기

연속간행물 권호 선택

연속간행물 권호 선택

우편복사 안내

도서위치안내(서울관)

저자프로필

목차보기

우편복사 안내

우편복사 목록담기

확인

내서재에 담기

새로운 서재

저장

로그인

권호