Journal of Korean Society of Agricultural Engineers ISSN:1738-3692(Print) 2093-7709(Online) 한국농공학회논문집

An increasing pattern of extreme rainfall recently affected the rural infrastructures with catastrophic damage, especially the overtopping of a fill dam embankment in the Republic of Korea. The overtopping was caused by the sudden increase in reservoir water level over the dam crest level, and it was not easy work to predict a priori because of its non-linear behavior. Fuzzy time series (FTS) is a fuzzy-logic inference procedure and is suited to apply to non-linear prediction methods such as machine learning. This study used the Wangshin reservoir and Goesan-dam cases, which experienced overtopping in 2023 and 2022, respectively. Wangshin Reservoir was a typical agricultural fill dam and needed to stack more available data, with only the daily storage rate (water level) of 7 years, starting on 2 May 2016. Therefore, we used Goesan-dam data to select appropriate variables and compare the analysis result, which was stacked with about 17 years of records. The analyses adapted LSTM to compare with FTS. As a result, the reservoir water level was applied to predict the overtopping water level, and it was shown that the FTS method could predict the actual water levels effectively according to the result of comparison with LSTM. Then, the FTS method was expected to predict reservoir water level a priori to make appropriate countermeasures on overtopping events as one of the alternatives.