J Korean Soc Hazard Mitig 2018; 18(1): 1-11  https://doi.org/10.9798/KOSHAM.2018.18.1.1
Prediction of River Water Level Using Deep-Learning Open Library
Jung, Sungho*, Lee, Daeeop**, and Lee, Kyoungsang***
*Member, Master’s Course, Dept. of Construction and Disaster Prevention Engineering, Kyungpook National University
**Member, Ph.D. Candidate, Dept. of Construction and Disaster Prevention Engineering, Kyungpook National University
***Member, Master’s Course, Dept. of Construction and Disaster Prevention Engineering, Kyungpook National University
Correspondence to: Member, Ph.D. Candidate, Dept. of Construction and Disaster Prevention Engineering, Kyungpook National University (Tel: +82-54-535-1360, Fax: +82-54-535-1360, E-mail: hydroeop@gmail.com)
Received: September 29, 2017; Revised: October 18, 2017; Accepted: November 20, 2017; Published online: January 31, 2018.
© The Korean Society of Hazard Mitigation. All rights reserved.

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Abstract
This research aims to predict the water level using deep-learning algorithm. To achieve this goal, we applied the TensorFlow, a deep-learning open source library to predict the water level of Okcheon station in the Guem-river. For model training and prediction, two hourly water level data sets of the 3 water level stations (Sutong, Hotan, Songcheon) are prepared: from 2002 to 2013 (training); from 2014 to 2016 (prediction). Even if many of physical data are necessary to understand water cycle system, in particular, model rainfall-runoff process, we used only upstream observed water level information to predict downstream water level using multi-linear regression and Long Short-Term Memory(LSTM) models based on the TensorFlow. The results showed that the weights(or regression coefficients) of multi-linear regression model were very fluctuated due to training trials, then, the predicted high water level were too much underestimated than the observations. On the other hand, the LSTM model predicted the downstream water level very stably regardless of water level height for the study period because sequence length of the LSTM memorize antecedent water level information for model training and updating.
Keywords: Deep-learning, TensorFlow, Multi Linear Regression, LSTM, Water Level


This Article

e-submission

Archives

Indexed/Covered by