Application of Long Short‒Term Memory neural network for time series prediction of flow rate at My Thuan hydrology station, Tien river

Abstract

Flow rate prediction has an important role in water resource management to reduce potential damage caused by floods for urban residential areas. However, prediction of flow rate presents great challenges because the task requires a number of information, such as hydrological, geomorphological data. The objective of this paper is to apply an effective approach for flow rate forecasting at My Thuan hydrology station (Tien River), based on the construction of a Long Short‒Term Memory (LSTM) neural network model using flow rate monitoring data These data composed of 8760 hourly flow rate data points within 2018. Coefficient of determination (R2), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE) are used to evaluate performances of LSTM model. The study evaluates the ability of LSTM algorithm to predict water flow with different number of neurons (1, 2, 3, 4) at different forecasting time: 1, 2, 3, 4, 5 hours ahead (t + 1, t + 2, t + 3, t + 4, t + 5, respectively). The research results indicated that the LSTM model with 3 neurons achieved a high performance for flow rate forecasting. When forecasting one hour ahead (t + 1), R2, RMSE, MAE reached 0.937, 2294.60, and 1738.33, respectively for training period, and was 0.884, 2655.66, and 2064.30, respectively for testing period. The findings of this study suggest that the LSTM model has promised as a potential tool in flow rate forecasting at the My Thuan and for other hydrology stations in Vietnam.