UTILIZATION OF SIMULATED DATA FOR DEVELOPMENT OF A DEEP LEARNING FLOOD PREDICTION MODEL AS SUPPORT OF DRAINAGE PUMP OPERATION

Abstract

 The purpose of our study is to create a deep-learing model for water levels that can support appropriate operations of drainage facilities in an agricultural lowland. This model may help make a guidline that describe an effective use of a deep learning approach with artificial datasets for water level during heavy rainfall events even if insufficient observed data are avariable in the lowland. The study has two steps. First, the artificial data of water levels were made by a hy-drolological model using three cases of rainfall amounts (100, 300, and 500 mm/h) during three days, eeach of whose has 1, 000 different patterns. Secondtly, our deep-learning model that employs a LSTM model utilized the artificial data as observed data to predict water levels. LSTM model predicted water levels in an hour in the lowland after an appropriate learning process. The results show that the predic-tions of three cases have good aggreement with the observed data except for un-learned events like ex-tremely heavy rainfalls. As a future work, a lead-tme extension of the model and a comaprison among other deep-learning architectures are required.