A Data-driven Machine Learning Approach for Reservoir Water Level Forecasting

Dams are large infrastructure essential for most ASEAN countries, serving to store water for agriculture, water supply and disasters mitigation, including flood and drought management. Safety conditions of those dams are also directly concerned by the dam operators and communities. One of the most common causes of dam failures is hydrological hazards, which occur when the stored water level in the upstream reservoir exceeds the capacity of the dam itself. To prevent that incident of dam failure, the Reservoir Water Level forecasting is also very crucial for the dam operators. In this paper, a Long Short-Term Memory network is applied for forecasting the Reservoir Water Level based on long term historical data collected by the real time monitoring system. The model is developed and tested in Python with necessary libraries. The best performance of the model shown by Root Mean Square Error, Correlation coefficient and Standard deviation for the next 5 days to 15 days forecasting are in the range of 0.2571-0.6389, 0.9945-0.9563 and 2.1942-2.1969, respectively. It is also shown that combining the daily rainfall information, the Root Mean Square Error, Correlation coefficient and Standard deviation of 15 days forecasting could be improved to 0.61607, 0.96015 and 2.1969, respectively.