DOI QR코드

DOI QR Code

다중 입력 딥러닝을 이용한 서리 발생 추정

Estimation of Frost Occurrence using Multi-Input Deep Learning

  • 김용석 (국립농업과학원 기후변화평가과) ;
  • 허지나 (국립농업과학원 기후변화평가과) ;
  • 김응섭 (국립농업과학원 기후변화평가과) ;
  • 심교문 (국립농업과학원 기후변화평가과) ;
  • 조세라 (국립농업과학원 기후변화평가과) ;
  • 강민구 (국립농업과학원 기후변화평가과)
  • Yongseok Kim (Climate Change Assessment Division, National Institute of Agricultural Sciences) ;
  • Jina Hur (Climate Change Assessment Division, National Institute of Agricultural Sciences) ;
  • Eung-Sup Kim (Climate Change Assessment Division, National Institute of Agricultural Sciences) ;
  • Kyo-Moon Shim (Climate Change Assessment Division, National Institute of Agricultural Sciences) ;
  • Sera Jo (Climate Change Assessment Division, National Institute of Agricultural Sciences) ;
  • Min-Gu Kang (Climate Change Assessment Division, National Institute of Agricultural Sciences)
  • 투고 : 2023.12.12
  • 심사 : 2024.02.23
  • 발행 : 2024.03.30

초록

본 연구에서는 딥러닝을 이용한 모형을 이용해서 우리나라 지역에 대한 서리 발생 예측 모형을 구축하였다. 딥러닝 모형의 학습 데이터로 다양한 기상인자들(최저기온, 풍속, 상대습도, 구름량, 강수량)을 사용하였으며, 기상인자들에 대한 통계적 분석 결과, 서리가 발생한 날과 서리가 발생하지 않은 날에 대해 각 요소별로 유의한 차이가 있는 것을 볼 수 있었다. 단일 딥러닝 모형 3가지와 다중 입력 딥러닝 모형 3가지를 이용하여 서리발생을 추정한 결과, 평균적으로 MLP가 가장 정확도가 낮았으며, LSTM, GRU 순으로 정확도가 높게 나타났고, 다중 입력 딥러닝 모형의 경우 3가지 모형이 거의 비슷한 결과가 나타났지만 그 중 평균적으로 GRU와 MLP를 이용한 모형이 가장 정확도가 높았다. 또한, 단일 딥러닝이 다중 입력 딥러닝에 비해 샘플에 따라 정확도 편차도 더 컸다. 이에 따라 결과적으로 단일 딥러닝 기반의 서리발생 예측 모형보다 다중 입력 딥러닝 기반의 서리발생 예측 모형이 안정성과 정확도와 재현율 측면에서 다소 우수한 것을 확인할 수 있었다.

In this study, we built a model to estimate frost occurrence in South Korea using single-input deep learning and multi-input deep learning. Meteorological factors used as learning data included minimum temperature, wind speed, relative humidity, cloud cover, and precipitation. As a result of statistical analysis for each factor on days when frost occurred and days when frost did not occur, significant differences were found. When evaluating the frost occurrence models based on single-input deep learning and multi-input deep learning model, the model using both GRU and MLP was highest accuracy at 0.8774 on average. As a result, it was found that frost occurrence model adopting multi-input deep learning improved performance more than using MLP, LSTM, GRU respectively.

키워드

과제정보

본 연구는 농촌진흥청 "신농업기후변화대응체계구축사업(과제번호: RS-2020-RD009396)"의 지원으로 수행되었습니다.

참고문헌

  1. Chen, J. L., H. B. Liu, W. Wu, and D. T. Xie, 2010: Estimation of monthly solar radiation from measured temperatures using support vector machines - A case study. Renewable Energy 36(2), 413-420.
  2. Chen, P., B. Zou, A. N. Belkacem, X. Lyu, X. Zhao, W. Yi, Z. Huang, J. Liang and C. Chen, 2022: An improved multi-input deep convolutional neural network for automatic emotion recognition. Frontiers in Neuroscience 16.
  3. Chun, J. A., H.-J. Lee, S.-H. Im, D. Kim, and S. Baek, 2021: Comparative assessment of frost event prediction models using logistic regression, random forest, and LSTM networks. Journal of Korea Water Resources Association 54(9), 667-680.
  4. Dua, N., S. N. Singh, and V. B. Semwal, 2021: Multiinput CNNGRU based human activity recognition using wearable sensors. Computing 103, 1461-1478.
  5. Han, J. H., J. J. Choi, U. Chung, K. S. Cho, and J. P. Chun, 2009: Frostfall forecasting in the naju pear production area based on discriminant analysis of climate data. Korean Journal of Agricultural and Forest Meteorology 11(1), 13-18. (in Korean with English abstract)
  6. Kim, H. K., and S. Kim, 2022: A study on frost prediction model using machine learning. The Korean Journal of Applied Statistics 35(4), 543-552. (in Korean with English abstract)
  7. Kim, Y., K.-M. Shim, M.-P. Jung, and I.-T. Choi, 2017: Study on the Estimation of Frost Occurrence Classification Using Machine Learning Methods. Korean Journal of Agricultural and Forest Meteorology 19(3), 86-92. (in Korean with English abstract)
  8. Kwon, Y. A., H. S. Lee. W. T. Kwon, and K. O. Boo, 2008: The weather characteristics of frost occurrence days for protecting crops against frost damage. Journal of the Korean Geographical Society 43(6), 824-842. (in Korean with English abstract)
  9. Noh, I.S., H.W. Doh, S.O. Kim, S.H. Kim, S.E. Shin, S.J. Lee, 2021: Machine Learning-Based Hourly Frost-Prediction System Optimized for Orchards Using Automatic Weather Station and Digital Camera Image Data. Atmosphere 12(7), 846.
  10. Noh, I., S.-J. Lee, S. Lee, S.-J. Kim, and S.-D. Yang. 2021: A High-Resolution (20 m) Simulation of Nighttime Low Temperature Inducing Agricultural Crop Damage with the WRF-LES Modeling System. Atmosphere 12(12), 1562.
  11. Robinson, C. and N. Mort,1996: A neural network solution to the problem of frost Prediction. UKACC International Conference on Control. Control. 136-139.
  12. Temeyer, B. R., W. A. Gallus Jr, K. A. Jungbluth, D. Burkheimer, and D. McCauley, 2003: Using an artificial neural network to predict parameters for frost deposition on Iowa bridgeways. Proceedings of the 2003 Mid-Continent Transportation Researchh Symposium, Iowa state University, Ames, IA, 2003.