Authors
1
Master's student, Water Resources Civil Engineering Department, Faculty of Civil Engineering, Architecture and Art, Research Science Unit, Islamic Azad University, Tehran, Iran
2
Assistant professor and faculty member of the Department of Civil Engineering, Department of Water Resources Management, Science and Research Unit, Tehran, Islamic Azad University, Tehran, Iran
3
associate Professor, Department of Civil Engineering, Faculty of Technology and Engineering, Vali Asr University (AJ), Rafsanjan, Rafsanjan, Iran
Abstract
plays an essential role in water resources management. Hydrological drought affects water allocation and hydropower generation. Therefore, hydrological drought prediction can help for optimal water resources management. In this paper, short-term hydrological drought was predicted using deep learning. Doing so, Standardized Hydrological Drought Index (SHDI) was calculated in one, three and six months periods. Then, 36 combinations and three different modes of input variables were considered to predict SHDI. In the next step, various optimization algorithms, including grasshopper optimization algorithm, salt swarm algorithm, biogeography-based optimization, and particle swarm optimization combined with ANN were used to predict SHDI. Here, 70% of the data were considered for training and the remaining 30% for test. Finally, the performance of deep learning (DNN) compared to the combined ANN-SSA, -BBO, -PSO and -GOA were discussed. The results of ANN compared to the combined models showed that the combined models performed better, but they are less compatible to DNN. In the DNN on one hand does not need to define the input combination, and on the other hand could reduce the value of RMSE to 0.32.
Keywords
)