Publications

Detailed Information

A Study on the Improvement of Basic Wind Speed Calculation Method Using Machine Learning : 머신러닝을 이용한 기본풍속 산정 개선에 관한 연구

DC Field Value Language
dc.contributor.advisor강현구-
dc.contributor.author이동혁-
dc.date.accessioned2021-11-30T01:56:52Z-
dc.date.available2021-11-30T01:56:52Z-
dc.date.issued2021-02-
dc.identifier.other000000165027-
dc.identifier.urihttps://hdl.handle.net/10371/175106-
dc.identifier.urihttps://dcollection.snu.ac.kr/common/orgView/000000165027ko_KR
dc.description학위논문 (석사) -- 서울대학교 대학원 : 공과대학 건축학과, 2021. 2. 강현구.-
dc.description.abstractBasic wind speed is an important factor in the design of wind loads on structures. Wind speed in a particular area is affected by various factors such as terrain, location, and distribution of buildings in that area. Therefore, it is quite difficult to estimate the wind speed of a particular area. Currently, KBC 2016 deals with this problem using basic wind speed map made from the observed data of the weather stations across the country. However, the basic wind speed map of KBC 2016 does not reflect the influence of the terrain. And there is an uncertainty of surface roughness classification, which distinguishes into four categories depending on the condition of the terrain. This study deals with method to improve these limitations through machine learning. Recently, the development of machine learning has been rapidly made in various fields, and this has allowed data to be approached from a new perspective. In particular, by learning meaningful features from images and using them in sub-problems or downstream tasks, problems can be solved in a different approach. In this study, the terrain features affecting wind speed are quantitatively evaluated through the learning of the features of satellite imagery. In addition, the wind speed calculation method is improved through various methods with machine learning and its effects are analyzed. In the results, the terrain features are learned from satellite imagery and the improved method of calculating wind speed is confirmed to have a smaller calculation error than the existing method. It is also confirmed that improved results are obtained even though experiments are conducted with relatively shallow neural network models and using simple data. Therefore, it is expected that the addition of more sophisticated models and other data that affect wind speeds provides better wind speed calculation.-
dc.description.abstract기본 풍속은 구조물의 풍하중 설계의 기본이 되는 중요한 요소이다. 특정한 지역의 풍속은 해당 지역의 지형, 위치 및 건물의 분포 등 다양한 요인들에 의해 복합적으로 영향을 받는다. 때문에 특정 지역의 기본 풍속을 정하는 것은 어려운 일이다. 현재 KBC 2016에서는 전국 기상관측소의 관측 데이터로부터 추정한 기본 풍속도를 통해 이를 다루고 있다. 그러나 KBC 2016의 기본풍속도는 지형의 영향력을 반영하지 못하거나, 지형의 상태에 따라 4가지로 분류한 지표면조도구분을 통해 다루는 한계가 있다. 본 연구에서는 이러한 한계를 머신러닝을 통해 정성적으로 개선하는 방법을 다룬다. 최근 들어 머신러닝의 급격한 발전은 다양한 분야에서 사용되며, 이를 통해 데이터를 새로운 관점에서 접근할 수 있게 되었다. 특히, 이미지에서 의미 있는 특징을 학습하여 다른 하위 문제에 사용함으로써 기존과 다른 접근 방식으로 문제를 다룰 수 있다. 본 연구에서는 위성사진의 특징 학습을 통해 풍속에 영향을 미치는 지형적 특징을 정량적으로 평가하였다. 또한 이를 활용하여 기존의 풍속 산정방식을 다양한 방법을 통해 개선하여, 그 효과를 분석하였다. 분석결과, 위성사진에서 지형적 특징을 충분히 학습할 수 있었으며, 개선된 풍속 산정 방식이 기존방식보다 예측의 오차가 작은 것을 확인하였다. 본 연구에서는 비교적 얕은 신경망 모델과 간단한 데이터로 실험을 진행하였지만, 정교한 모델 및 풍속에 영향을 미치는 다른 데이터를 추가하였을 때 더 좋은 예측을 할 것으로 기대된다.-
dc.description.tableofcontentsAbstract i
Contents iii
List of Tables vi
List of Figures vii
Chapter 1. Introduction 1
1.1 Introduction 1
1.1.1 Basic wind speed 1
1.1.2 Machine learning 2
1.1.3 Satellite imagery 3
1.2 Purpose 4
1.3 Organization 7
Chapter 2. Background and Previous Studies 8
2.1 Basic Wind Speed 8
2.1.1 Concepts on wind speed 8
2.1.2 KBC 2016 code 9
2.1.3 Preview studies 15
2.2 Machine Learning 18
2.2.1 History of machine learning 18
2.2.2 Representation learning and self-supervised learning 20
2.2.3 Pre-trained and transfer learning 20
2.2.4 Convolutional Neural Network (CNN) 21
2.2.5 Variational AutoEncoder (VAE) 28
2.2.6 Representation learning 33
2.3 Satellite Imagery 36
2.3.1 Landsat 36
2.3.2 Machine learning using satellite imagery 37
Chapter 3. Baseline Experiment 38
3.1 Purpose of Baseline Experiment 38
3.2 Dataset and Preprocessing 39
3.2.1 Wind speed data 39
3.2.2 Wind speed data preprocessing 39
3.2.3 Heuristic preprocessing of data 47
3.3 Baseline Experiment 48
3.3.1 Experiment design 48
3.3.2 Experiment results 51
Chapter 4. Terrain Feature Learning 61
4.1 General 61
4.2 Preprocessing Method of Satellite Imagery 62
4.2.1 Cloudless RGB satellite imagery 63
4.2.2 Hillshade imagery 63
4.2.3 Image segmentation 64
4.3 Machine Learning Model to Learn Satellite Feature 67
4.3.1 Beta-TC VAE 68
4.3.2 SimCLR 74
4.4 Discussion 81
Chapter 5. Basic Wind Speed Estimation with Machine Learning 82
5.1 General 82
5.2 Rebalance Surface Roughness Using Machine Learning 83
5.2.1 VAE based 84
5.2.2 SimCLR based 85
5.2.3 Rebalance results 86
5.2.4 Discussion 90
5.3 Method Reflecting Terrain Similarity 91
5.3.1 Method using terrain similarity only 91
5.3.2 Method combining terrain similarity and distance 92
5.3.3 Discussion 95
5.4 End-to-End Method 96
5.4.1 Support Vector Regression (SVR) 96
5.4.2 Simple Multilayer Perceptron (MLP) 97
5.4.3 Discussion 97
Chapter 6. Conclusion 98
References 101
국 문 초 록 105
-
dc.format.extentix, 117-
dc.language.isoeng-
dc.publisher서울대학교 대학원-
dc.subjectBasic wind speed-
dc.subjectmachine learning-
dc.subjectsatellite imagery-
dc.subjectterrain feature-
dc.subject기본풍속-
dc.subject머신러닝-
dc.subject위성사진-
dc.subject지형특징-
dc.subject.ddc690-
dc.titleA Study on the Improvement of Basic Wind Speed Calculation Method Using Machine Learning-
dc.title.alternative머신러닝을 이용한 기본풍속 산정 개선에 관한 연구-
dc.typeThesis-
dc.typeDissertation-
dc.contributor.AlternativeAuthorDonghyeok Lee-
dc.contributor.department공과대학 건축학과-
dc.description.degreeMaster-
dc.date.awarded2021-02-
dc.embargo.liftdate2023-03-01-
dc.contributor.major건축구조-
dc.identifier.uciI804:11032-000000165027-
dc.identifier.holdings000000000044▲000000000050▲000000165027▲-
Appears in Collections:
Files in This Item:
There are no files associated with this item.

Altmetrics

Item View & Download Count

  • mendeley

Items in S-Space are protected by copyright, with all rights reserved, unless otherwise indicated.

Share