Skip to content

This app is a solution that predicts the water stage(level) of bridges across Yuseong-gu.

License

Notifications You must be signed in to change notification settings

enfycius/Miri-Su

Folders and files

NameName
Last commit message
Last commit date

Latest commit

Β 

History

33 Commits
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 

Repository files navigation

미리수(水)

Abstract

Β Recently, due to abnormal climate conditions, South Korea has experienced an increase in the frequency of heavy rainfall, leading to an increase in instances of river overflow and low-lying area inundation. However, predicting the locations and timing of floods due to the variability of climate conditions is a challenging task that requires a substantial amount of data and complex calculations. Thus, in this study, we utilized data from the Geumgang Flood Control Office spanning from 2017 to 2021 to predict the water levels at the Gapcheon Mannyeon Bridge in 2022. Furthermore, in this study, we investigated the interrelationship among precipitation, volume flow rate, and water levels in order to achieve low error rates with a limited amount of data. Both the Prophet and LightGBM models were employed in machine learning, and the accuracy of predicted river water levels was evaluated using metrics such as Mean Absolute Percentage Error(MAPE) and Root Mean Square Error (RMSE). The evaluation of the trained models revealed that both the Prophet and LightGBM models demonstrated a low level of error in predicting river water levels. On average, the MAPE value was 3.02%, the MSE value was 0.43%, and the RMSE value was 6.49%. These results suggest that the developed models can be effectively used to predict river water levels, and this study highlights an intuitive and simple approach to predicting water levels in the Gapcheon area.Also, it provides guidelines for creating a water level prediction model with a low error rate, including when the amount of data to be learned is small.

Key Words: Machine Learning, Water Level, Precipitation, Volume Flow Rate, Correlation

Introduction

 졜근 μ΄μƒκΈ°ν›„λ‘œ 인해 μ „ μ„Έκ³„μ—μ„œ λ‹€μ–‘ν•œ ν˜•νƒœμ˜ μžμ—°μž¬ν•΄λ‘œ μΈν•œ μ‹¬κ°ν•œ ν”Όν•΄κ°€ 보고되고 μžˆλ‹€. 특히, μ΄λŸ¬ν•œ μžμ—°μž¬ν•΄ 쀑 호우둜 μΈν•œ μ €μ§€λŒ€ 침수 ν˜„μƒμ€ 인λͺ…, μž¬μ‚° 및 μ‚¬νšŒ 경제 μ‹œμŠ€ν…œμ— λŒ€κ·œλͺ¨λ‘œ ν”Όν•΄λ₯Ό μ•ΌκΈ°ν•˜λ©° κ°€μž₯ 파괴적인 영ν–₯을 λ―ΈμΉœλ‹€. λ˜ν•œ, κ·Έ 영ν–₯은 더 이상 μ§€μ—­μ μ΄κ±°λ‚˜ μ‹œκ°„μ μœΌλ‘œ μ œν•œλ˜μ§€ μ•Šκ³  μ „ 세계적인 문제둜 μ§„ν™”ν•˜κ³  μžˆλ‹€.

 특히, ν•œκ΅­μ²˜λŸΌ 인ꡬ 밀도가 높은 μ§€μ—­μ—μ„œλŠ” 호우둜 μΈν•œ μΉ¨μˆ˜κ°€ λ”μš± μ‹¬κ°ν•œ 문제둜 λΆ€μƒν•˜κ³  μžˆλ‹€. ꡭ가적 μ°¨μ›μ—μ„œλ„ μ΄λŸ¬ν•œ λ¬Έμ œμ— μ§λ©΄ν•˜κ³  μžˆμŒμ—λ„ λΆˆκ΅¬ν•˜κ³  λ„μ‹œ ν•˜μ²œκ³Ό 같은 인ꡬ λ°€μ§‘μ§€μ—­μ—μ„œμ˜ λ²”λžŒμ€ λŒ€κ·œλͺ¨ ν™μˆ˜ ν”Όν•΄λ‘œ 이어지고 μžˆλ‹€. 졜근의 μ‚¬λ‘€λ‘œλŠ” 2020λ…„ ν•œλ°˜λ„ 폭우 μ‚¬νƒœμ™€ 2022λ…„ μ€‘λΆ€κΆŒ 폭우 μ‚¬νƒœκ°€ μžˆλ‹€. μ΄λŸ¬ν•œ 사둀듀은 ν™μˆ˜λ‘œ μΈν•œ ν•˜μ²œμ˜ λ²”λžŒμ„ μ‹œμž‘μœΌλ‘œ μ‚°μ‚¬νƒœ, μ €μ§€λŒ€ 침수 λ“± λ‹€μ–‘ν•œ ν˜•νƒœμ˜ ν”Όν•΄κ°€ μ—°μ‡„μ μœΌλ‘œ λ°œμƒν–ˆλ‹€.

Β κΈ°ν›„ λ³€ν™”(Climate Change)둜 μΈν•œ ν™μˆ˜μ˜ ν”Όν•΄λ₯Ό μ˜ˆλ°©ν•˜κ³  μ΅œμ†Œν™”ν•˜κΈ° μœ„ν•΄μ„œλŠ” μˆ˜λ¬Έν•™μ  관점(Hydrological Perspective)μ—μ„œμ˜ 단기 및 μž₯기적인 예츑이 맀우 μ€‘μš”ν•˜κΈ° λ•Œλ¬Έμ— λ‹€μ–‘ν•œ 예츑 λͺ¨λΈμ— λŒ€ν•œ 연ꡬ가 ν™œλ°œνžˆ μ§„ν–‰λ˜κ³  μžˆλ‹€(Kim et al., 2022; Kim et al., 2022; Jung et al., 2017; Osman et al., 2018; Taylor and Letham., 2017). κ·ΈλŸ¬λ‚˜ ν™μˆ˜ λ°œμƒ μœ„μΉ˜μ™€ μ‹œκΈ°μ— λŒ€ν•œ μ˜ˆμΈ‘μ€ κΈ°ν›„ 쑰건의 동 적 μ„±μ§ˆ λ•Œλ¬Έμ— 근본적으둜 λ³΅μž‘ν•˜λ‹€. λ”°λΌμ„œ 였늘 λ‚  μ£Όμš”ν•œ ν™μˆ˜ 예츑 λͺ¨λΈμ€ 데이터에 νŠΉν™”λ˜μ–΄ λ‹¨μˆœν™”λœ 가정듀이 ν¬ν•¨λ˜λ©°, 물리적 κ³Όμ •κ³Ό 볡작 ν•œ μˆ˜ν•™μ  식을 λͺ¨λ°©ν•˜κΈ° μœ„ν•΄ 결정둠적, κ²½ν—˜μ , ν™•λ₯ μ  기법듀을 ν™œμš©ν•˜μ—¬ 이루어진닀. 물리 기반 λͺ¨λΈμ€ 폭풍(Costabile and Macchione., 2012; Costabile and Macchione., 2013), ν•˜μ²œ 및 ν•΄μ–‘(Le et al., 2019; Borah., 2011)κ³Ό 같은 μˆ˜λ¬Έν•™μ  관점에 μ„œ μ˜ˆμΈ‘ν•˜λŠ”λ° μ‚¬μš©λ˜μ—ˆλ‹€. κ·ΈλŸ¬λ‚˜ 물리 λͺ¨λΈμ€ λ‹€ μ–‘ν•œ 수문 μ‹œμŠ€ν…œμ„ μ˜ˆμΈ‘ν•˜κΈ° μœ„ν•΄μ„œ λ§Žμ€ 데이터 λ₯Ό ν™•λ³΄ν•˜λŠ”λ° ν•œκ³„κ°€ 있으며, κ³„μ‚°μ˜ λ³΅μž‘μ„± λ•Œλ¬Έ 에 μ˜ˆμΈ‘μ— ν•„μš”ν•œ μˆ˜λ¬Έν•™μ  맀개 λ³€μˆ˜λ“€μ„ κ³ λ €ν•˜ 기에 μ–΄λ €μ› λ‹€.

Β ν™μˆ˜ 예츑 λͺ¨λΈ κ°œλ°œμ„ μœ„ν•œ 데이터 확보와 계 μ‚°μ˜ λ³΅μž‘μ„±κ³Ό 같은 ν•œκ³„λ₯Ό κ·Ήλ³΅ν•˜κΈ° μœ„ν•΄, μš°λ¦¬λŠ” μ œν•œλœ μ‹œκ³„μ—΄ 데이터λ₯Ό ν™œμš©ν•˜μ—¬ Prophet λͺ¨λΈκ³Ό LightGBMκ³Ό 같은 κΈ°κ³„ν•™μŠ΅(Machine Learning) μ•Œκ³  λ¦¬μ¦˜μ„ κ΅¬μΆ•ν–ˆλ‹€. μš°λ¦¬κ°€ μ§λ©΄ν•œ μ£Όμš” κ³Όμ œλŠ” μ œν•œ 된 μ‹œκ³„μ—΄ 데이터λ₯Ό ν™œμš©ν•΄ κΈ°κ³„ν•™μŠ΅(Machine Learning)λͺ¨λΈμ„ ν•™μŠ΅μ‹œν‚€κ³  κ°‘μ²œμ˜ μˆ˜μœ„λ₯Ό μ˜ˆμΈ‘ν•˜ κΈ° μœ„ν•œ λ‹¨μˆœλ²•(Simple Method)을 μ°ΎλŠ” 것이닀. 특 히 μš°λ¦¬λŠ” λ°μ΄ν„°μ˜ 뢀쑱함을 κ·Ήλ³΅ν•˜κ³  λͺ¨λΈμ˜ μ • 확도λ₯Ό ν–₯μƒμ‹œν‚€λŠ”λ° μ΄ˆμ μ„ λ§žμΆ”κ³  μžˆλ‹€. 이λ₯Ό μœ„ ν•΄ 2017λ…„λΆ€ν„° 2021λ…„κΉŒμ§€μ˜ κ°•μˆ˜λŸ‰(Precipitation), μˆ˜μœ„(Water Level), μ²΄μ μœ λŸ‰(Volume Flow Rate)의 μƒν˜Έ 연관성을 νŒŒμ•…ν•˜μ—¬ μ΅œμ’…μ μœΌλ‘œ 미래의 νŠΉμ • μ‹œμ μ˜ μˆ˜μœ„λ₯Ό μ˜ˆμΈ‘ν–ˆλ‹€. μ˜ˆμΈ‘ν•œ μˆ˜μœ„μ˜ μ„±λŠ₯ 평가 λ₯Ό μœ„ν•΄μ„œ μš°λ¦¬λŠ” 평균 μ ˆλŒ€ λ°±λΆ„μœ¨ 였차(Mean Absolute Percentage Error, MAPE), 평균 제곱 였차 (Mean Sqaure Error, MSE), 평균 제곱근 였차(Root Mean Square Error, RMSE) μ„Έ 가지λ₯Ό 평가 μ§€ν‘œλ‘œ μ„ μ •ν•˜μ˜€λ‹€. 이 μ§€ν‘œλ“€μ„ 톡해 μ‹€μ œ μˆ˜μœ„μ™€ 예츑된 μˆ˜μœ„ κ°„μ˜ 였차λ₯Ό μ •λŸ‰ν™”ν•˜κ³ , λͺ¨λΈμ˜ 예츑 μ„±λŠ₯을 ν‰κ°€ν–ˆλ‹€.

Overview

Β λ³Έ μ—°κ΅¬μ˜ κ΅¬μ„±μš”μ†Œ 및 κ·Έ μˆœμ„œλ„(Flow Chart)λŠ” λ‹€μŒ Fig. 1κ³Ό κ°™λ‹€.

Figure 1

Fig. 1. Flow Chart for Building Flood Prediction Method

Β λ³Έ μ—°κ΅¬μ—μ„œλŠ” μœ λŸ‰ 및 κ°•μˆ˜λŸ‰ 자료λ₯Ό ν†΅ν•΄μ„œ 침수 μ˜ˆλ°©μ„ μœ„ν•˜μ—¬ μˆ˜μœ„ 예츑 μ‹œμŠ€ν…œμ„ κ°œλ°œν•˜μ˜€λ‹€. 이λ₯Ό μœ„ν•˜μ—¬ μœ λŸ‰κ³Ό κ°•μˆ˜λŸ‰μ— λŒ€ν•œ μ‹œκ³„μ—΄ 자료λ₯Ό μˆ˜μ§‘ν•˜μ˜€μœΌλ©°, 결츑치 제거 λ“±μ˜ μ „μ²˜λ¦¬λ₯Ό μ§„ν–‰ν–ˆλ‹€. λ‹€μŒμœΌλ‘œ, 인곡지λŠ₯ ν•™μŠ΅ κ³Όμ •μ—μ„œ 과적합을 막기 μœ„ν•˜μ—¬ 2017λ…„λΆ€ν„° 2022λ…„μ˜ 전체 μ‹œκ³„μ—΄ 데이터듀을 ν›ˆλ ¨μš©(2017λ…„ ~ 2021λ…„ 12μ›”)κ³Ό κ²€μ¦μš©(2022λ…„ 01μ›” ~ 2022λ…„ 12μ›”)으둜 λΆ„ν• ν•˜μ—¬ ν•™μŠ΅ μ‹œ 과적합 μ—¬λΆ€λ₯Ό ν™•μΈν•˜μ˜€λ‹€. κ·Έλ‹€μŒ, Prophetκ³Ό LightGBM λͺ¨λΈμ„ κ΅¬μΆ•ν•˜κ³ , κ°€μ€‘μΉ˜λ₯Ό μž„μ˜λ‘œ μ΄ˆκΈ°ν™”ν•˜μ—¬ λͺ¨λΈμ„ ν›ˆλ ¨μ‹œμΌ°μœΌλ©°, λͺ¨λΈμ˜ μ„±λŠ₯을 ν™•μΈν•˜κΈ° μœ„ν•˜μ—¬ 평균 μ ˆλŒ€ λ°±λΆ„μœ¨ 였차(Mean Absolute Percentage Error, MAPE)와 평균 제곱 였차(Mean Square Error, MSE) 평균 제곱근 였차(Root Mean Square Error, RMSE)의 μ„Έ 개의 μ„±λŠ₯μ§€ν‘œλ₯Ό μ΄μš©ν•˜μ˜€λ‹€. μ΄λŸ¬ν•œ μ„±λŠ₯μ§€ν‘œλ₯Ό 톡해 λͺ¨λΈμ˜ 정확도λ₯Ό ν‰κ°€ν•œ λ‹€μŒ, 두 λͺ¨λΈ 쀑 정확도가 더 λ†’κ²Œ μ‚°μΆœλ˜λŠ” λͺ¨λΈμ„ 톡해 μ‹€μ œ ν˜„μƒμΈ κ°‘μ²œμ˜ λ§Œλ…„κ΅ μˆ˜μœ„μ— λŒ€ν•œ μ˜ˆμΈ‘μ„ μ§„ν–‰ν•˜λ„λ‘ λ³Έ 연ꡬλ₯Ό κ΅¬μ„±ν•˜μ˜€λ‹€.

Data Collection, Analysis, and Preprocessing

Data Collection

Β λ³Έ μ—°κ΅¬μ—μ„œλŠ” λŒ€μ „κ΄‘μ—­μ‹œ κ°‘μ²œμ˜ ν™μˆ˜μ˜ˆλ°© 및 μž¬λ°œλ°©μ§€λ₯Ό μœ„ν•˜μ—¬ λ§Œλ…„κ΅λ₯Ό μ£Ό μ—°κ΅¬λŒ€μƒμœΌλ‘œ μ„ μ •ν•˜μ˜€λ‹€. κΈ°κ³„ν•™μŠ΅(Machine Learning) λͺ¨λΈ ꡬ좕에 μ•žμ„œ, μˆ˜μœ„λ₯Ό μ˜ˆμΈ‘ν•  수 μžˆλŠ” νŠΉμ§•(Feature)에 ν•΄λ‹Ήν•˜λŠ” μžλ£Œμ™€ μ •λ‹΅(Labels)인 μˆ˜μœ„μžλ£Œμ˜ μˆ˜μ§‘μ„ μœ„ν•˜μ—¬ κΈˆκ°• ν™μˆ˜ν†΅μ œμ†Œμ—μ„œ μ œκ³΅ν•˜λŠ” 2017λ…„ 1μ›”λΆ€ν„° 2022λ…„ 12μ›”κΉŒμ§€ 총 6λ…„μ˜ μœ λŸ‰ 및 κ°•μˆ˜λŸ‰, μˆ˜μœ„ 데이터λ₯Ό ν™œμš©ν•˜μ˜€λ‹€.

Figure 2

Fig. 2. Geographical Location of Gabcheon and Enlarged Photograph of the Mannyeon Bridge

Β Fig. 2λŠ” 침수 예츑 λͺ¨λΈμ˜ ꡬ좕을 μœ„ν•΄ μˆ˜μ§‘ν•œ μ‹œκ³„μ—΄ 데이터λ₯Ό μ‚¬μš©ν•œ λŒ€μƒ 지역인 κ°‘μ²œμ˜ 지리적 νŠΉμ§•κ³Ό ν™•λŒ€λœ λ§Œλ…„κ΅μ˜ 사진을 보여쀀닀. κ°‘μ²œμ€ λŒ€λ‘”μ‚°μ—μ„œ μ‹œμž‘ν•˜μ—¬ λŒ€μ „ 지역을 ν†΅κ³Όν•˜λ©° 길이가 73.7km인 κΈˆκ°• 지λ₯˜μ˜ ν•œ 뢀뢄이닀. 이 κ°‘μ²œμ€ κΈˆκ°•κ³Ό ν•©λ₯˜ν•˜κΈ° 전에 λŒ€μ „ 지역을 κ°€λ‘œμ§€λ₯΄λ©° μœ μ—­ λ‚΄ μ§€μ—­μ˜ 수질 및 μˆ˜λŸ‰ 변화에 μ€‘μš”ν•œ 영ν–₯을 λ―ΈμΉœλ‹€. κ°‘μ²œμ˜ μˆ˜μœ„λŠ” κ³„μ ˆμ— 따라 λ³€λ™ν•˜λŠ”λ°, λ΄„μ—λŠ” Melting Snow와 λΉ—λ¬Όλ‘œ 인해 μˆ˜μœ„κ°€ μƒμŠΉν•  수 있고, μ—¬λ¦„μ—λŠ” μž₯λ§ˆμ™€ κ°•μš°λ‘œ 인해 μˆ˜μœ„κ°€ λ”μš± 높아진닀.

 특히, λ§Œλ…„κ΅λŠ” κ°‘μ²œ μ£Όλ³€μ—μ„œ λ°œμƒν•˜λŠ” κ°•μˆ˜λŸ‰ μ¦κ°€λ‘œ 인해 μž₯마 κΈ°κ°„ λ™μ•ˆ 반볡적으둜 μΉ¨μˆ˜λ˜λŠ” μ£Όμš” 지점 쀑 ν•˜λ‚˜μ΄λ‹€. μ—°λ‘€μ μœΌλ‘œ ν™μˆ˜λ‘œλΆ€ν„° 영ν–₯을 λ°›μ•„ 지역 μ‚¬νšŒμ™€ ꡐ톡에 λΆˆνŽΈμ„ μ΄ˆλž˜ν•˜λŠ” 사둀가 λ°œμƒν•˜κ³  μžˆλ‹€. λ”°λΌμ„œ 우리 μ—°κ΅¬λŠ” κ°‘μ²œμ—μ„œ μˆ˜μ§‘ν•œ λ‹€μ–‘ν•œ μš”μ†Œμ˜ μ‹œκ³„μ—΄ 데이터λ₯Ό ν™œμš©ν•œ 예츑 λͺ¨λΈμ„ κ°œλ°œν•˜μ—¬ ν™μˆ˜ λ°œμƒ κ°€λŠ₯성을 사전에 μ˜ˆμΈ‘ν•˜κ³ μž ν•œλ‹€.

Data Analysis

κ°•μˆ˜λŸ‰(Precipitation)

Figure 3

Fig. 3. Visualization of the Distribution of Precipitation in Gapcheon from 2017 to 2021

Β Fig. 3은 2017λ…„λΆ€ν„° 2021λ…„κΉŒμ§€μ˜ κ°‘μ²œμ˜ κ°•μˆ˜λŸ‰μ˜ 뢄포λ₯Ό μ‹œκ°ν™”ν•œ κ·Έλž˜ν”„μ΄λ‹€. 이 κ·Έλž˜ν”„μ—μ„œ νŒŒλž€μ λ“€μ€ 각 μ‹œκ°„λ‹Ή κ°•μˆ˜λŸ‰μ„ λ‚˜νƒ€λ‚΄λ©°, 이λ₯Ό 톡해 연도별 κΈ°ν›„ 변화와 κ°•μˆ˜λŸ‰ 뢄포λ₯Ό νŒŒμ•…ν•  수 μžˆλ‹€.

Β Fig. 3μ—μ„œ λ³Ό 수 μžˆλ“―μ΄, λͺ¨λ“  λ…„λ„μ—μ„œ 3λΆ„κΈ°(여름~가을)에 μ‹œκ°„λ‹Ή κ°•μˆ˜λŸ‰μ΄ 높은 κ²½ν–₯을 보이며, μ΄λŠ” κ°‘μ²œ μ§€μ—­μ˜ κΈ°ν›„ νŠΉμ„±μ„ 확인할 수 μžˆλ‹€. 특히, 2018λ…„κ³Ό 2020λ…„μ—λŠ” 각각 70mm/hour와 58mm/hour둜 3뢄기에 높은 μ‹œκ°„λ‹Ή κ°•μˆ˜λŸ‰μ„ κΈ°λ‘ν–ˆλ‹€. μ΄λŸ¬ν•œ 값듀은 λ‹€λ₯Έ 년도와 비ꡐ해도 μƒλ‹Ήνžˆ 높은 μˆ˜μ€€μ— ν•΄λ‹Ήν•˜λ©°, 두 ν•΄μ˜ 3λΆ„κΈ°λŠ” κΈ°ν›„μ μœΌλ‘œ 비ꡐ적 μŠ΅ν•œ 쑰건을 κ°€μ§€λ˜ μ‹œκΈ°λ‘œ ν•΄μ„λœλ‹€.

μˆ˜μœ„(Water Stage)

Figure 4

Fig. 4. Visualization of the Distribution of Water Stage in Gapcheon from 2017.01 to 2021.12

Β Fig. 4λŠ” 2017λ…„λΆ€ν„° 2021λ…„κΉŒμ§€μ˜ κ°‘μ²œμ˜ μˆ˜μœ„ 뢄포λ₯Ό μ‹œκ°ν™”ν•œ κ·Έλž˜ν”„μ΄λ‹€. 이 κ·Έλž˜ν”„μ—μ„œ νŒŒλž€μ„ λ“€μ€ μ‹œκ°„μ— λ”°λ₯Έ κ°‘μ²œμ˜ μˆ˜μœ„ λ³€ν™”λ₯Ό λ‚˜νƒ€λ‚΄λ©°, μ—°λ„λ³„λ‘œ μˆ˜μœ„μ˜ 변동을 νŒŒμ•…ν•  수 μžˆλ‹€. 특히, Fig. 3μ—μ„œ λ³Ό 수 μžˆλ“―μ΄, 2018λ…„κ³Ό 2020λ…„μ˜ 3λΆ„κΈ°μ—λŠ” 높은 κ°•μˆ˜λŸ‰μ΄ κ΄€μ°°λ˜μ—ˆλ‹€. μ΄λ‘œμΈν•΄ κ°‘μ²œμ˜ μˆ˜μœ„κ°€ κΈ‰κ²©ν•˜κ²Œ μƒμŠΉν•˜λŠ” κ²½ν–₯을 λ³΄μ˜€μœΌλ©°, μ΄λŠ” κ°•μˆ˜λŸ‰κ³Ό μˆ˜μœ„ 사이에 μƒν˜Έ 연관성이 μžˆμŒμ„ μœ μΆ”ν•΄λ³Ό 수 μžˆλ‹€.

 결과적으둜, 2018λ…„κ³Ό 2020년에 κ΄€μ°°λœ 높은 κ°•μˆ˜λŸ‰κ³Ό μˆ˜μœ„μ˜ μƒμŠΉμ€ μ„œλ‘œ κ°•ν•œ 연관성을 λ‚˜νƒ€λ‚΄λ©°, 특히 3뢄기에 μ΄λŸ¬ν•œ κ²½ν–₯성이 λ‘λ“œλŸ¬μ Έ 보인닀. μ΄λŸ¬ν•œ 관계성은 κ°‘μ²œμ—μ„œμ˜ κ°•μš°μ™€ μˆ˜μœ„ μƒμŠΉκ°„μ˜ 관계λ₯Ό λͺ…ν™•ν•˜κ²Œ 보여주며, μˆ˜μœ„ λ³€ν™”κ°€ κ°•μš°μ˜ 영ν–₯을 크게 λ°›λŠ”λ‹€λŠ” 것을 λ§ν•œλ‹€.

μ²΄μ μœ λŸ‰(Volume Flow Rate)

Figure 5

Fig. 5. Visualization of the Distribution of Volume Flow Rate in Gapcheon from 2017.01 to 2021.12

Β Fig. 5λŠ” 2017λ…„λΆ€ν„° 2021λ…„κΉŒμ§€μ˜ κ°‘μ²œμ˜ μ²΄μ μœ λŸ‰μ˜ 뢄포λ₯Ό μ‹œκ°ν™”ν•œ κ·Έλž˜ν”„μ΄λ‹€. 이 κ·Έλž˜ν”„μ—μ„œ νŒŒλž€μ„ λ“€μ€ μ‹œκ°„μ— λ”°λ₯Έ μ²΄μ μœ λŸ‰μ„ μ˜λ―Έν•˜λ©°, 이λ₯Ό 톡해 연도별 κ°‘μ²œμ§€μ—­μ˜ μ²΄μ μœ λŸ‰μ„ νŒŒμ•„ν•  수 μžˆμ—ˆλ‹€. 특히, 2018λ…„κ³Ό 2020λ…„μ—λŠ” μ²΄μ μœ λŸ‰μ΄ 각각 μ•½ 570 m3/s와 1200m3/s둜, λ‹€λ₯Έ 연도와 λΉ„κ΅ν•˜μ—¬ 높은 μˆ˜μ€€μ˜ μ²΄μ μœ λŸ‰μ„ λ³΄μ˜€λ‹€.

 이에 따라 κ°•μˆ˜λŸ‰ 증가(Fig. 3 μ°Έμ‘°)κ°€ μˆ˜μœ„μ˜ 증가(Fig. 4 μ°Έμ‘°)둜 이어지며, μ¦κ°€ν•œ κ°‘μ²œμ˜ 체적으둜 인해 μ²΄μ μœ λŸ‰ λ˜ν•œ 증가(Fig. 5 μ°Έμ‘°)ν•˜λŠ” μ–‘μƒμœΌλ‘œ 이어진 것을 확인할 수 μžˆλ‹€. λ˜ν•œ, κ°•μˆ˜λŸ‰, μˆ˜μœ„ 그리고 μ²΄μ μœ λŸ‰μ˜ 접점(Peak Point)의 지점듀이 λ™μΌν•œ κ²½ν–₯을 λ³΄μΈλ‹€λŠ” 사싀도 κ΄€μ°°λœλ‹€. μ΄λŠ” 높은 κ°•μˆ˜λŸ‰μ„ κΈ°λ‘ν•œ μ‹œκΈ°μ— κ°‘μ²œμ˜ μˆ˜μœ„κ°€ μ¦κ°€ν•˜μ—¬ μ²΄μ μœ λŸ‰μ΄ μ¦κ°€ν•˜λŠ” κ²½ν–₯성을 λ‚˜νƒ€λ‚΄λ©°, μ΄λŠ” κ°•μˆ˜λŸ‰, μˆ˜μœ„ 그리고 μ²΄μ μœ λŸ‰ μ‚¬μ΄μ˜ 연관성을 νŒŒμ•…ν•˜λŠ” 것이 κ°‘μ²œμ˜ μˆ˜μœ„ 예츑 λͺ¨λΈμ„ κ°œλ°œν•˜λŠ” 데 μ€‘μš”ν•œ μš”μ†ŒμΌ 것이라 νŒλ‹¨λœλ‹€.

Data Preprocessing

결츑치(Missing Values) 제거

Figure 6

Fig. 6. Missing Values of Core Features

Β Fig. 6κ³Ό 같이 2018λ…„κ³Ό 2019년에 일뢀 데이터가 μ—†λŠ” 것을 확인할 수 μžˆλ‹€. λ”°λΌμ„œ 결츑치λ₯Ό μ±„μš°κΈ° μœ„ν•΄μ„œ λ³΄κ°„λœ κ°’μœΌλ‘œ μ±„μ›Œλ„£λŠ” 보간법을 μ‚¬μš©ν•˜μ˜€λ‹€.

Figure 7

Fig. 7. Filled by Interpolated Values

Β Fig. 7은 보간법을 μ μš©ν•˜μ—¬ 결츑치λ₯Ό μ±„μ›Œλ„£μ€ λͺ¨μŠ΅μ΄λ‹€.

λ¦¬μƒ˜ν”Œλ§(Resampling)

Β λ¦¬μƒ˜ν”Œλ§(Resampling)을 μ§„ν–‰ν•˜μ—¬ 데이터λ₯Ό μ¦λŒ€ λ˜λŠ” κ°μ†Œμ‹œν‚¬ 수 μžˆλ‹€. λ³Έ μ—°κ΅¬μ—μ„œλŠ” 24μ‹œκ°„μœΌλ‘œ λ‚˜λˆ„μ–΄μ§„ 데이터λ₯Ό ν‰κ· κ°’μœΌλ‘œ λŒ€μ²΄ν•˜μ—¬ 일 λ‹¨μœ„λ‘œ λŠμ–΄μ„œ ν•™μŠ΅μ„ μ§„ν–‰ν•˜μ˜€λ‹€.

정상성(Stationarity)

 정상성(Stationarity)은 λ°μ΄ν„°μ˜ ν™•λ₯ μ μΈ μ„±μ§ˆλ“€μ΄ μ‹œκ°„μ˜ 흐름에 따라 λ³€ν™”ν•˜μ§€ μ•ŠλŠ”λ‹€λŠ” 것을 μ˜λ―Έν•œλ‹€. μ΄λŠ” 평균, 곡뢄산, λΆ„μ‚° 등이 μ‹œκ°„μ— μ˜μ‘΄ν•˜μ§€ μ•ŠλŠ”λ‹€λŠ” 것을 μ˜λ―Έν•œλ‹€. μš°λ¦¬κ°€ μ‚¬μš©ν•˜λŠ” 데이터가 κ³Όμ—° 비정상성을 κ°–κ³  μžˆλŠ”μ§€ 확인해야 좔세와 κ³„μ ˆμ„±μ„ 가진 μ‹œκ³„μ—΄ λ°μ΄ν„°λ‘œ 뢄석할 의미λ₯Ό κ°–λŠ”λ‹€. 정상성은 μ‹œκ°μ , 기초 ν†΅κ³„λŸ‰, 톡계적 검정을 ν†΅ν•΄μ„œ 확인할 수 μžˆλ‹€. κ·Έ 쀑 Augmented Dickey - Fuller (ADF) 검정은 λ‹¨μœ„κ·Ό κ²€μ •μœΌλ‘œ 비정상성을 확인할 수 μžˆλ‹€.

  • 귀무가섀 H0: μ‹œκ³„μ—΄μ€ κ³ μ •λ˜μ–΄ μžˆμ§€ μ•Šλ‹€.
  • λŒ€λ¦½κ°€μ„€ H1: μ‹œκ³„μ—΄μ€ 정지 μƒνƒœμ΄λ‹€.

 귀무가섀을 κΈ°κ°ν•˜λŠ” 방법은 두 κ°€μ§€λ‘œ P-valueλ₯Ό ν†΅ν•œ 방법과 ADF ν†΅κ³„λŸ‰μ„ ν†΅ν•œ 방법이 μžˆλ‹€. P-valueλ₯Ό ν†΅ν•œ 방법은 P-value > μœ μ˜μˆ˜μ€€(Default: 0.05)이면 귀무가섀을 μ±„νƒν•œλ‹€. P-value <= μœ μ˜μˆ˜μ€€(Default: 0.05)이면 귀무가섀을 κΈ°κ°ν•œλ‹€. ADF ν†΅κ³„λŸ‰μ„ ν†΅ν•œ 방법은 ADF Statistics > Critical Value이면 귀무가섀을 μ±„νƒν•œλ‹€. ADF Statistics < Critical Value이면 귀무가섀을 κΈ°κ°ν•œλ‹€.

Figure 8

Fig. 8. P-values & ADF Statistics of Volume Flow Rate

Β Fig. 8μ—μ„œ μ²΄μ μœ λŸ‰μ˜ ADF Statistics의 값은 -5.088으둜 Critical Values 5%에 ν•΄λ‹Ήν•˜λŠ” -2.871보닀 큰 값이 λ„μΆœλ˜μ—ˆλ‹€. λ”°λΌμ„œ μ²΄μ μœ λŸ‰μ˜ μ‹œκ³„μ—΄ μžλ£ŒλŠ” κ³ μ •λ˜μ–΄ μžˆμ§€ μ•Šλ‹€.

Figure 9

Fig. 9. P-values & ADF Statistics of Precipitation

Β Fig. 9μ—μ„œ κ°•μˆ˜λŸ‰μ˜ ADF Statistics의 값은 -11.797으둜 Critical Values 5%에 ν•΄λ‹Ήν•˜λŠ” -2.871보닀 큰 값이 λ„μΆœλ˜μ—ˆλ‹€. λ”°λΌμ„œ κ°•μˆ˜λŸ‰ μ‹œκ³„μ—΄ μžλ£ŒλŠ” κ³ μ •λ˜μ–΄ μžˆμ§€ μ•Šλ‹€.

Figure 10

Fig. 10. P-values & ADF Statistics of Water Stage

Β Fig. 10μ—μ„œ μˆ˜μœ„μ˜ ADF Statistics의 값은 -8.188으둜 Critical Values 5%에 ν•΄λ‹Ήν•˜λŠ” -2.871보닀 큰 값이 λ„μΆœλ˜μ—ˆλ‹€. λ”°λΌμ„œ μˆ˜μœ„ μ‹œκ³„μ—΄ μžλ£ŒλŠ” κ³ μ •λ˜μ–΄ μžˆμ§€ μ•Šλ‹€.

Β λ‹€μŒκ³Ό 같이 값이 νŠΉμ • λ³€μˆ˜λ“€μ˜ μ‹œκ³„μ—΄μ΄ κ³ μ •λ˜μ–΄ μžˆμ§€ μ•ŠμŒμ„ 보여쀀닀. 이λ₯Ό ν•΄κ²°ν•˜λŠ” λ°©λ²•μœΌλ‘œ 데이터λ₯Ό λ³€ν™˜ν•˜μ—¬ μ μš©μ„ ν•΄μ•Όν•˜λŠ”λ°, 일반적인 두 가지 방법은 둜그 λ˜λŠ” μ œκ³±κ·Όμ„ μ μš©ν•˜κ±°λ‚˜ μ°¨λΆ„ν•˜μ—¬ μ‹œκ³„μ—΄μ„ κ³ μ •ν•  수 μžˆλ‹€. λ³Έ μ—°κ΅¬μ—μ„œλŠ” μ°¨λΆ„ν•˜μ—¬ μ‹œκ³„μ—΄μ„ κ³ μ •ν•˜λŠ” 방법을 μ‚¬μš©ν•˜μ˜€λ‹€. μ΄λŸ¬ν•œ 차뢄은 Eq. 1의 1μ°¨ μ°¨λΆ„κ³Ό Eq. 2의 2μ°¨ μ°¨λΆ„μœΌλ‘œ λ‚˜λ‰œλ‹€.

Β  1μ°¨ 차뢄은 λ‹€μŒ 식을 μ μš©ν•œλ‹€.

$$z_{i} = y_{i} - y_{i-1}$$

Eq. 1. 1st Difference Equation

Eq. 1의 1μ°¨ 차뢄을 각 μžλ£Œμ— μ μš©ν•œ κ²°κ³Ό κ·Έλž˜ν”„λŠ” Fig. 11 ~ 13κ³Ό κ°™λ‹€.

Figure 11

Fig. 11. First-order Differentiated Volume Flow Rate

Figure 12

Fig. 12. First-order Differentiated Precipitation

Figure 13

Fig. 13. First-order Differentiated Water Stage

Β  2μ°¨ 차뢄은 λ‹€μŒ 식을 μ μš©ν•œλ‹€.

$$z_{i} = (y_{i} - y_{i-1}) - (y_{i-1} - y_{i-2})$$

Eq. 2. 2nd Difference Equation

Eq. 2의 2μ°¨ 차뢄을 각 μžλ£Œμ— μ μš©ν•œ κ²°κ³Ό κ·Έλž˜ν”„λŠ” Fig. 14 ~ 16κ³Ό κ°™λ‹€.

Figure 14

Fig. 14. Second-order Differentiated Volume Flow Rate

Figure 15

Fig. 15. Second-order Differentiated Precipitation

Figure 16

Fig. 16. Second-order Differentiated Water Stage

κ·Έ κ²°κ³Ό, 1μ°¨ 차뢄을 μ μš©ν•œ κ²°κ³Ό κ·Έλž˜ν”„μ—μ„œ 더 적은 정상성을 가짐을 ν™•μΈν•˜μ˜€κ³ , 이에 2μ°¨ 차뢄이 μ•„λ‹Œ, 1μ°¨ μ°¨λΆ„μœΌλ‘œ μ „μ²˜λ¦¬λ₯Ό μ§„ν–‰ν•˜μ˜€λ‹€.

히트맡(Heat Map)

Β μˆ˜μ§‘ν•œ μœ λŸ‰ 및 κ°•μˆ˜λŸ‰ 데이터와 μˆ˜μœ„ κ°„μ˜ 상관관계λ₯Ό ν™•μΈν•˜κΈ° μœ„ν•˜μ—¬ 상관관계(Correlations)에 λŒ€ν•œ ν–‰λ ¬(Matrix)을 μƒμ„±ν•œ λ‹€μŒ, 이λ₯Ό 히트 맡(Heat Map) ν˜•νƒœλ‘œ κ·Έ 관계λ₯Ό λ„μ‹ν™”ν•˜μ˜€μœΌλ©°, κ·Έ κ²°κ³ΌλŠ” λ‹€μŒ Fig. 17κ³Ό κ°™μ•˜λ‹€.

Figure 17

Fig. 17. Visualization of Heat Maps for Correlations between Features and Labels

Β  Fig. 17λ‘œλΆ€ν„° 히트 맡의 λŒ€κ°μ„±λΆ„λ“€μ€ λͺ¨λ‘ λ™μΌν•œ νŠΉμ§•λ“€μ— λŒ€ν•œ μƒκ΄€κ΄€κ³„μ΄λ―€λ‘œ 1이 λ„μΆœλ˜μ—ˆμœΌλ©°, μ²΄μ μœ λŸ‰(Volume Flow Rate)κ³Ό μˆ˜μœ„(Water Stage), κ°•μˆ˜λŸ‰(Precipitation)κ³Ό μˆ˜μœ„(Water Stage)간에 각각 0.84, 0.87둜 νŠΉμ§•(Features)κ³Ό μ •λ‹΅(Labels)λ“€ 간에 비ꡐ적 μ„ ν˜•μ˜ 상관관계λ₯Ό λ³΄μž„μ„ ν™•μΈν•˜μ˜€λ‹€.

ν›ˆλ ¨(Train) 및 ν…ŒμŠ€νŠΈ(Test) 데이터셋(Datasets)

Table. 1. Train and Test Datasets

Date
(2017.01.01)
Volume Flow Rate
(m3/s)
Precipitation
(mm/hr)
Water Stage
(m)
(=y)
00:00:00 2.68 0.0 1.07
01:00:00 2.84 0.0 1.08
02:00:00 2.84 0.0 1.08
$\cdots$ $\cdots$ $\cdots$ $\cdots$

Β  Table. 1μ—μ„œ 확인할 수 μžˆλ“―μ΄, 데이터셋(Datasets)은 기본적으둜 λ‹€λ³€λŸ‰(Multivariate)의 μ‹œκ³„μ—΄(Time Series) λ°μ΄ν„°λ‘œ, μ²΄μ μœ λŸ‰(Volume Flow Rate)κ³Ό κ°•μˆ˜λŸ‰(Precipitation)의 두 νŠΉμ§• 데이터와 정닡인 μˆ˜μœ„(Water Stage) λ°μ΄ν„°λ‘œ κ΅¬μ„±λ˜μ–΄ μžˆλ‹€. 예츑 λͺ¨λΈλ‘œμ˜ 데이터 μ£Όμž…μ΄ μš©μ΄ν•˜λ„λ‘ 기쑴에 λΆ„λ¦¬λ˜μ–΄ 있던 μ„Έ 개의 κ°œλ³„μ μΈ 데이터듀을 ν•˜λ‚˜μ˜ 자료 ν˜•νƒœλ‘œ μ·¨ν•©ν•˜μ˜€λ‹€.

Build and Train Models

Prophet Model

Β  μˆ˜μœ„ μ˜ˆμΈ‘μ„ μœ„ν•œ λ‹€λ³€λŸ‰ κΈ°κ³„ν•™μŠ΅ λͺ¨λΈλ‘œμ„œ 페이슀뢁(Facebook)μ—μ„œ κ³΅κ°œν•œ μ‹œκ³„μ—΄ 예츑 λͺ¨λΈμΈ Prophet을 μ„ μ •ν•˜μ˜€λ‹€. λ³Έ λͺ¨λΈμ€ μΌλ°˜ν™”λœ κ°€μ‚° λͺ¨λΈ(GAM)(Hastie & Tibshirani 1987)κ³Ό μœ μ‚¬ν•˜λ©°, νšŒκ·€λͺ¨λΈμž„에도 λΆˆκ΅¬ν•˜κ³  μ‹œκ°„μ„ κ΅¬μ„±μš”μ†Œλ‘œ ν•˜λŠ” μ„ ν˜•κ³Ό λΉ„μ„ ν˜• νšŒκ·€ λͺ¨λΈ κΈ°λŠ₯을 μˆ˜ν–‰ν•  수 μžˆλ‹€. GAM의 ν˜•νƒœλ₯Ό κ°–μΆ€μœΌλ‘œ μƒˆλ‘œμš΄ κ΅¬μ„±μš”μ†Œλ₯Ό μΆ”κ°€ν•˜κ³ , λ³€ν˜•ν•˜κΈ° 쉽닀. ARIMA λͺ¨λΈκ³Ό 달리 츑정값에 μΌμ •ν•œ 간격을 μœ μ§€ν•  ν•„μš”κ°€ μ—†μœΌλ©°, λˆ„λ½λœ 값을 보간할 ν•„μš”κ°€ μ—†λ‹€. ν•™μŠ΅μ˜ 속도도 λ›°μ–΄λ‚˜ 더 μ‰½κ²Œ λͺ¨λΈμ„ μ‚¬μš©ν•  수 μžˆλ‹€.

Eq. 3. Prophet Model

$$y(t) = g(t) + s(t) + h(t) + \epsilon_{t}$$

Β  Prophet λͺ¨ν˜•μ€ Eq. 3와 같은 ν•¨μˆ˜λ₯Ό 가지며, μΆ”μ„Έ(Trend), κ³„μ ˆμ„±(Seasonality), 휴일과 이벀트(Holidays and Events)의 3가지 μ£Όμš” μš”μ†Œλ‘œ κ΅¬μ„±λœλ‹€.

μΆ”μ„Έ(Trend)

Β  비주기적 λ³€ν™”λ₯Ό λͺ¨λΈλ§ν•˜λŠ” μΆ”μ„Έ ν•¨μˆ˜λ‘œ 기본적으둜 μ„±μž₯성을 λ°˜μ˜ν•œλ‹€.

Eq. 4. Piecewise Logistic Growth Model

$$g(t) = \frac{C}{1 + exp(-k(t-m))}$$

Β  μ—¬κΈ°μ„œ $C$λŠ” ν•œκ³„μ μ„ λ‚˜νƒ€λ‚΄λŠ” 수용λ ₯, $k$λŠ” μ„±μž₯λ₯ , $t$와 $m$은 각각 μ‹œκ°„κ³Ό μ˜€ν”„μ…‹(Offset) νŒŒλΌλ―Έν„°λ₯Ό μ˜λ―Έν•œλ‹€.

κ³„μ ˆμ„±(Seasonality)

Β  주기적 λ³€ν™”λ₯Ό λ°˜μ˜ν•˜λŠ” ν•¨μˆ˜λ‘œ λ°˜λ³΅λ˜λŠ” νŒ¨ν„΄(Pattern)에 νŠΉμ§•μ„ λ°˜μ˜ν•œλ‹€.

Eq. 5. Seasonality Approximation Fourier Function

$$s(t) = \sum_{i=1}^{N}(a_{n}cos(\frac{2\pi{nt}}{P}) + b_{n}sin(\frac{2\pi{nt}}{P}))$$

Β  μ—¬κΈ°μ„œ $P$λŠ” μ‹œκ³„μ—΄ λͺ¨λΈμ—μ„œ κΈ°λŒ€ν•˜λŠ” μ •κ·œμ£ΌκΈ°λ‘œ κΈ°κ°„(Periods)을 μ˜λ―Έν•œλ‹€.

Applications

🚧 Architectures

User Interface (UI)

UI

Fig. x. User Interface

Demo

Demo

Fig. x. Demo Alerts

References

Kim, D., Han, H., Wang, W., & Kim, H. S. (2022). Improvement of Deep Learning Models for River Water Level Prediction Using Complex Network Method. Water, 14(3), 466.

Kim, D., Lee, K., Hwang-Bo, J. G., Kim, H. S., & Kim, S. (2022). Development of the Method for Flood Water Level Forecasting and Flood Damage Warning Using an AI-based Model. Journal of the Korean Society of Hazard Mitigation, 22(4), 145–156.

Kyungpook National University, Jung, S., Lee, D., Kyungpook National University, Lee, K., & Kyungpook National University. (2017). Prediction of River Water Level Using Deep-Learning Open Library. Korean Society of Hazard Mitigation, 18(1), 1–11.

Osman, S., Aziz, N. A., Husaif, N., Sidek, L. M., Shakirah, A., Hanum, F., & Basri, H. (2018). Application of Stochastic Flood Forecasting Model Using Regression Method for Kelantan Catchment. MATEC Web of Conferences, 203, 07001.

Taylor, S. J., & Letham, B. (2017). Forecasting at scale[Preprint]. PeerJ Preprints.

Costabile, P., Costanzo, C., & Macchione, F. (2012). Comparative analysis of overland flow models using finite volume schemes. Journal of Hydroinformatics, 14(1), 122–135.

Costabile, P., Costanzo, C., & Macchione, F. (2013). A storm event watershed model for surface runoff based on 2D fully dynamic wave equations: A WATERSHED RUNOFF MODEL BASED ON 2D FULLY DYNAMIC WAVE EQUATIONS. Hydrological Processes, 27(4), 554–569.

Le, Ho, Lee, & Jung. (2019). Application of Long Short-Term Memory (LSTM) Neural Network for Flood Forecasting. Water, 11(7), 1387.

Borah, D. K. (2011). Hydrologic procedures of storm event watershed models: A comprehensive review and comparison. Hydrological Processes, 25(22), 3472–3489.

Comments

λ³Έ κΈ€κ³Ό μ €μž₯μ†Œ(Repository)의 μ €μž‘κΆŒμ€ κΉ€μ’…ν˜, 이성지, 김민쀑, λ‚˜μ’…μ—΄μ—κ²Œ μžˆμŠ΅λ‹ˆλ‹€. 무단 λ„μš©, 배포, 볡제λ₯Ό κΈˆν•©λ‹ˆλ‹€.

λ˜ν•œ, μ•±μ˜ λ””μžμΈ 및 μ†ŒμŠ€μ½”λ“œλŠ” λ°•μœ μƒ, κ³ νƒœμ™„, κΉ€μ’…ν˜μ—κ²Œ μžˆμŠ΅λ‹ˆλ‹€. 무단 λ„μš©, 배포, 볡제λ₯Ό κΈˆν•©λ‹ˆλ‹€.

ν•„μš”ν•˜μ‹  μžλ£Œλ‚˜ 정보가 μžˆμœΌμ‹  경우, λ³Έ λ…Όλ¬Έμ˜ κ΅μ‹ μ €μž(Corresponding Author)인 λ‚˜μ’…μ—΄μ˜ μ΄λ©”μΌλ‘œ μ—°λ½μ£Όμ‹œκΈ° λ°”λžλ‹ˆλ‹€.

About

This app is a solution that predicts the water stage(level) of bridges across Yuseong-gu.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published