Reservoir volume forecasting using artificial intelligence-based models: Artificial Neural Networks, Support Vector Regression, and Long Short-Term Memory

Abstract

Reservoirs are essential structures that have important functions in water supply, flood control, irrigation for agriculture, and hydroelectric production. These functions can only be realized with an effective reservoir operation. Artificial intelligence (AI) techniques can be used for forecasting reservoir volumes, which is a critical parameter for reservoir operation and management. This study predicts the reservoir volumes using AI techniques, Artificial Neural Networks (ANN), Support Vector Regression (SVR), and Long Short-Term Memory (LSTM), in two reservoirs (Ladik and Yedikir Reservoirs). The effects of various parameters on the performance of the models were analyzed. The results show that (1) the number of epochs affects the performance of the ANN and LSTM models considerably, but the SVR model performance is sensitive to kernel function type, (2) the SVR model does not produce as good results when compared to the ANN and LSTM models. The SVR model performance can be improved with more advanced strategies, (3) the LSTM model, which is good at modelling time series data, showed the best performance for both reservoirs, indicating that it is more adaptive to temporal dynamics in water volumes than the ANN and SVR models. Overall, AI-based models provided promising results for estimating reservoir volumes.