Development of Modified LSTM Model for Reservoir Capacity Prediction in Huanggang Reservoir, Fujian, China

Abstract

The Huanggang Reservoir capacity is affected by a variety of factors. In order to accurately understand the Huanggang Reservoir capacity change, we develop a new hydrological prediction model based on the LSTM (Long-Short-Term Memory) method, which is used to predict the capacity of the reservoir. In this modified model, we choose to input multidimensional factors, two fully connected layers, selecting the optimal number of the hidden neurons, the optimizer, and adding the attention mechanism. The result of using the Developed LSTM and usual LSTM shows that the prediction curve of the Developed LSTM model can fit the true value better than the usual LSTM model, and the mean relative error of the Developed LSTM model decreased by 1.15%-3.82%, comparing with the usual LSTM model. Thus, we realize that the Developed LSTM model can make accurately prediction in some reservoir capacity estimations.