서울시 대기 중 미세먼지 농도 예측을 위한 3D 심층 합성곱 신경망 앙상블 예측 = Ensemble Prediction of 3D Deep Convolutional Neural Networks for the Prediction of PM2.5 Level for the Seoul, Republic of Korea

국문 초록 (Abstract)

대기 중 미세먼지 (Particular Matter)의 높은 농도에 노출되는 것은 인간 건강에 심각한 영향을 미친다.
많은 연구가 미세먼지가 특히 높은 농도에서 인간에게 매우 유독하다는 것을 보여주었다. 이러한 이유로 여러 국가가 미세먼지 농도를 규제하기 위해 상당한 노력을 기울이고 있다. 미세먼지 피해를 줄이기 위해 PM 농도 기반의 조기 경보 시스템이 필요다. 본 논문에서 대한민국 서울의 PM2.5 농도를 예측하기 위해 3D CNN을 활용한새로운 앙상블 예측 접근법을 제안한다. 이 방법은 매 6시간 최대 2일 동안 관측된 데이터에 대한 시공간 데이터의 특징을 추출하고 결합하기 위해 공간 정보뿐만 아니라 시간 정보도 활용할 수 있는 3D CNN의 앙상블 예측을사용한다. 결합한 특징은 공유 계층을 통해 의미 있는 특징 정보를 추출하고 개별 계층을 통해 시간별로 예측된 농도를 반환한다. 제안된 방법은 서울 메트로폴리탄 지역의 PM2.5 및 기상 데이터를 사용하여 2015년 1월 1일부터2021년 2월 28일까지 예측을 수행하였다. 제안된 방법은 PM 2.5 관측 (실제 값), CMAQ 예측 및 ConvLSTM에대해 종합적으로 모델을 평가하였다. 예측 성능 측면에서, 제안된 방법은 널리 사용되는 예측 모델과 비교하여7.06%의 향상된 예측 정확도, 4.28%의 향상된 고농도 탐지율, 오경보율을 11.77%만큼 개선하였다.

다국어 초록 (Multilingual Abstract)

Exposure to high concentrations of airborne particulate matter (Particular Matter) has serious effects on human health. Much research work showed that particulate matter is very toxic to humans, especially in high concentrations. For this reason, many countries make considerable efforts to regulate PM concentrations. In order to implement PM measures to reduce damage, an early warning system based on PM concentration level is essentially required. In this paper, we propose a novel 3D CNN ensemble prediction approach for forecasting PM2.5 concentrations in Seoul, Republic of Korea, which is observed every 6 hours for up to two days. The proposed method uses an ensemble of 3D deep CNNs that can utilize not only spatial information but also temporal information to extract and combine features for spatiotemporal data corresponding to each input. The combined features aim at extracting meaningful feature information through the shared layer and return the time-series predicted concentration for each period through the individual layer. The proposed method performs prediciton using PM2.5 and meteorological data of Seoul metropolitan area, ranging from 2015-01-01 to 2021-02-28. The proposed method was comprehensively evaluated for PM2.5 observation (Ground-truth), CMAQ forecast, and ConvLSTM. In terms of prediction performance, the proposed method showed improvement in performances with an increase of 7.06% in accuracy, an increase of 4.28% in probability of detection, and a decrease of 11.77% in false alram rate, compared to the widely used forecast model.

참고문헌 (Reference)

1 United States Environmental Protection Agency, "What is PM, and how does it get into the air? Particulate Matter(PM)Basics"

2 K. Simonyan, "Very deep convolutional networks for large-scale image recognition"

3 F. Lu, "Systematic review and meta-analysis of the adverse health effects of ambient PM25 and PM10 pollution in the Chinese population" 136 : 196-204, 2015

4 N. A. H. Janssen, "Short-term effects of PM2. 5, PM10 and PM2. 5–10 on daily mortality in the Netherlands" 463 : 20-26, 2013

5 J. E. Nash, "River flow forecasting through conceptual models part I—A discussion of principles" 10 (10): 282-290, 1970

6 D. Byun, "Review of the governing equations, computational algorithms, and other components of the Models-3Community Multiscale Air Quality(CMAQ)modeling system" 59 (59): 51-77, 2006

7 J. Meng, "Raster data projection transformation based-on Kriging interpolation approximate grid algorithm" 60 (60): 2013-2019, 2021

8 T. Yasuno, "Rain-code fusion: Code-to-code ConvLSTM forecasting spatiotemporal precipitation. In Pattern recognition" Springer International Publishing 20-34, 2021

9 J. L. Devore, "Probability and statistics for engineering and the sciences (Vol. 5)" Duxbury Press 1995

10 W. Z. Lu, "Prediction of pollutant levels in causeway bay area of Hong Kong using an improved neural network model" 128 (128): 1146-1157, 2002

11 M. L. Scapellato, "Personal PM10 exposure in asthmatic adults in Padova, Italy : Seasonal variability and factors affecting individual concentrations of particulate matter" 212 (212): 626-636, 2009

12 Environmental Performance Index, "PM2.5 exposure. RESULTS"

13 D. Kang, "New categorical metrics for air quality model evaluation" 46 (46): 549-555, 2007

14 P. M. Bartier, "Multivariate interpolation to incorporate thematic surface data using inverse distance weighting (IDW)" 22 (22): 795-799, 1996

15 M. C. Turner, "Long-term ambient fine particulate matter air pollution and lung cancer in a large cohort of never-smokers" 184 (184): 1374-1381, 2011

16 J. Schmidhuber, "Long short-term memory" 9 (9): 1735-1780, 1997

17 Air Korea, "Introduction to the Comprehensive air-quality index(CAI). What’s CAI"

18 L. S. Chang, "Human-model hybrid Korean air quality forecasting system" 66 (66): 896-911, 2016

19 K. Chakraborty, "Forecasting the behavior of multivariate time series using neural networks" 5 (5): 961-970, 1992

20 A. K. Paschalidou, "Forecasting hourly PM 10 concentration in Cyprus through artificial neural networks and multiple regression models : Implications to local environmental management" 18 : 316-327, 2011

21 S. Wu, "Fine particulate matter, temperature, and lung function in healthy adults : Findings from the HVNR study" 108 : 168-174, 2014

22 Air Korea, "Emergency measures to reduce fine dust"

23 H. Mahmoudabadi, "Directional kriging implementation for gridded data interpolation and comparative study with common methods" 2016 : G21B-1006, 2016

24 J. B. Lee, "Development of a deep neural network for predicting 6 h average PM 2.5 concentrations up to 2 subsequent days using various training data" 15 (15): 3797-3813, 2022

25 H. Chang-Hoi, "Development of a PM2.5 prediction model using a recurrent neural network algorithm for the Seoul metropolitan area, Republic of Korea" 245 : 118021-, 2021

26 D. W. Byun, "Development and implementation of the EPA’s models-3 initial operating version:Community multi-scale air quality (CMAQ)model" 7 : 357-368, 1998

27 Y. S. Lin, "Constructing PM2.5 map based on mobile PM2.5 sensor and cloud platform" IEEE 702-707, 2016

28 W. C. Skamarock, "A time-split nonhydrostatic atmospheric model for weather research and forecasting applications" 227 (227): 3465-3485, 2008

29 H. C. Chen, "A novel prediction approach for exploring PM2. 5 spatiotemporal propagation based on convolutional recursive neural networks"

30 G. Yang, "A hybrid deep learning model to forecast particulate matter concentration levels in Seoul, South Korea" 11 (11): 348-, 2020

31 T. C. Bui, "A deep learning approach for forecasting air pollution in South Korea using LSTM"

32 C. J. Huang, "A deep CNN-LSTM model for particulate matter (PM2.5) forecasting in smart cities" 18 (18): 2220-, 2018

33 R. Borge, "A comprehensive sensitivity analysis of the WRF model for air quality applications over the Iberian Peninsula" 42 (42): 8560-8574, 2008