Optimization of physical and hyperparameters for enhancing the response performance of an LSTM-based virtual flood level sensor

Yunji Lim; Donggyun Kim; Yeongeun Ji; Jiyoung Sung; Boosik Kang

doi:10.3741/JKWRA.2025.58.3.225

All Issue

2025 Vol.58, Issue 3 Preview Page Next Page

Research Article

Optimization of physical and hyperparameters for enhancing the response performance of an LSTM-based virtual flood level sensor LSTM기반 가상 홍수위 센서의 응답성능 강화를 위한 물리 및 하이퍼파라미터 최적화

31 March 2025. pp. 225-239

PDF XML

Abstract

Efficient river management and flood response require an increase in monitoring density. However, physical sensors installed on bridges are limited by high costs and placement constraints. To address these challenges, this study proposes a flood level prediction model utilizing a software-based virtual sensor. The proposed virtual sensor achieves measurement accuracy comparable to that of traditional physical sensors. This study developed an LSTM (Long Short-Term Memory) model using data collected from river observation stations and smart poles in the Seomjin River basin to learn long-term dependencies and nonlinear patterns in time series data. Furthermore, by optimizing the configuration of training data and prediction points, the model achieved high accuracy in predicting downstream water levels using upstream data. Through an optimized data pipeline, the virtual sensor was able to continuously predict water levels even after the removal of physical smart poles, providing real-time responsiveness for faster and more precise river management. This study demonstrates the potential of virtual sensors and highlights their practical contributions to river management and flood forecasting.

Keywords

Virtual sensor

LSTM

Flood water level prediction

Data pipeline design

효율적인 하천 관리와 홍수 대응을 위해 관측 밀도의 증대는 필수적이다. 그러나 교량에 설치되는 물리적 센서는 높은 비용과 설치 위치의 제약으로 인해 한계가 존재한다. 이러한 문제를 해결하고자 본 연구는 소프트웨어 기반 가상센서를 활용한 수위 예측 모델을 제안한다. 제안된 가상센서는 기존 물리 센서와 유사한 계측 정확도를 달성할 수 있다. 본 연구는 섬진강 유역의 하천 관측소와 스마트폴에서 수집된 데이터를 활용하여 시계열 데이터의 장기적 의존성과 비선형적 패턴을 학습하는 LSTM (Long Short-Term Memory) 모델을 개발하였다. 또한, 학습 데이터의 구성과 예측 지점 설정을 최적화함으로써 상류 데이터를 기반으로 하류 수위를 예측할 때 높은 정확도를 달성하였다. 최적화된 데이터 파이프라인을 통해 스마트폴이 제거된 이후에도 지속적으로 수위를 예측할 수 있는 가상센서를 구현하였으며, 이를 통해 실시간 응답 성능을 제공하여 더욱 신속하고 정밀한 하천 관리가 가능할 것으로 기대된다. 본 연구는 가상센서의 활용 가능성을 제시하며, 하천 관리 및 홍수 예보 분야에서 실질적 기여를 할 수 있을 것으로 보인다.

키워드

가상센서

인공지능

장단기메모리

홍수위 예측

데이터 파이프라인 설계

References

Atashi, V., Gorji, H.T., Shahabi, S.M., Kardan, R., and Lim, Y.H. (2022). "Water level forecasting using deep learning time- series analysis: A case study of Red River of the North." Water, Vol. 14, 1971.

10.3390/w14121971

Bengio, Y., Simard, P., and Frasconi, P. (1994). "Learning long-term dependencies with gradient descent is difficult." IEEE Transactions on Neural Networks, Vol. 5, No. 2, pp. 157-166.

10.1109/72.27918118267787

Bergstra, J., and Bengio, Y. (2012). "Random search for hyper-parameter optimization." Journal of Machine Learning Research, Vol. 13, pp. 281-305.

Cheong, T.S., Choi, C., Ye, S.J., Shin, J., Kim, S., and Koo, K.M. (2023). "Development of flood early warning frameworks for small streams in Korea." Water, Vol. 15, 1808.

10.3390/w15101808

Depuru, S.S.S.R., Depuru, S.S.R., and Depuru, S. (2023). "IoT-based smart water level monitoring." Proceedings of the 2023 IEEE 20th India Council International Conference (INDICON), Hyderabad, India, pp. 1136-1142.

10.1109/INDICON59947.2023.10440795

Hochreiter, S., and Schmidhuber, J. (1997). "Long short-term memory." Neural Computation, Vol. 9, No. 8, pp. 1735-1780.

10.1162/neco.1997.9.8.17359377276

Hu, C., Wu, Q., Li, H., Jian, S., Li, N., and Lou, Z. (2018). "Deep learning with a long short-term memory networks approach for rainfall-runoff simulation." Water, Vol. 10, 1543.

10.3390/w10111543

Kim, J.H. (2019). Application of FLOW-3D for simulating river hydraulics and sediment transport processes. Ph. D. Dissertation, University of Alberta, Edmonton, Alberta, Canada, pp. 15-20.

Li, W., Kiaghadi, A., and Dawson, C. (2020). "Exploring the best sequence LSTM modeling architecture for flood prediction." Neural Computing and Applications, Vol. 33, No. 11, pp. 5571-5580.

10.1007/s00521-020-05334-3

Lo, S.W., Wu, J.H., Lin, F.P., and Hsu, C.H. (2015). "Visual sensing for urban flood monitoring." Sensors, Vol. 15, pp. 20006- 20029.

10.3390/s15082000626287201PMC4570407

Moradi, E., Yaghoubi, B., and Shabanlou, S. (2023). "A new technique for flood routing by nonlinear Muskingum model and artificial gorilla troops algorithm." Applied Water Science, Vol. 13, 49.

10.1007/s13201-022-01844-8

Ponce, V.M., and Yevjevich, V. (1978). "Muskingum-Cunge method with variable parameters." Journal of the Hydraulics Division, ASCE, Vol. 104, No. 12, pp. 1663-1667.

10.1061/JYCEAJ.0005119

Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., and Salakhutdinov, R. (2014). "Dropout: A simple way to prevent neural networks from overfitting." Journal of Machine Learning Research, Vol. 15, pp. 1929-1958.

Tawalbeh, R., Alasali, F., Ghanem, Z., Alghazzawi, M., Abu-Raideh, A., and Holderbaum, W. (2023). "Innovative characterization and comparative analysis of water level sensors for enhanced early detection and warning of floods." Journal of Low Power Electronics and Applications, Vol. 13, 26.

10.3390/jlpea13020026

Wu, Z., Huang, Y., Huang, K., Yan, K., and Chen, H. (2023). "A review of non-contact water level measurement based on computer vision and radar technology." Water, Vol. 15, 3233.

10.3390/w15183233

Information

Publisher :KOREA WATER RESOURECES ASSOCIATION
Publisher(Ko) :한국수자원학회
Journal Title :Journal of Korea Water Resources Association
Journal Title(Ko) :한국수자원학회 논문집
Volume : 58
No :3
Pages :225-239
Received Date : 2024-11-11
Revised Date : 2025-01-31
Accepted Date : 2025-02-03
DOI :https://doi.org/10.3741/JKWRA.2025.58.3.225

Journal of Korea Water Resources Association ISSN:2799-8746(Print) 2799-8754(Online) 한국수자원학회 논문집

All Issue