1. Investigating hybrid deep learning models and meta-heuristic algorithms in predicting evaporation from a reservoir: a case study of Dez dam.
- Author
-
Farzad, Reza, Ahmadi, Farshad, Sharafati, Ahmad, and Hosseini, Seyed Abbas
- Subjects
- *
METAHEURISTIC algorithms , *DEEP learning , *SHORT-term memory , *LONG-term memory , *DAMS , *FLOOD control , *RESERVOIRS - Abstract
Reservoirs are crucial for water storage, flood control, and electricity generation. At the same time, evaporation in dam reservoirs causes water losses. This study evaluates the effect of weather conditions on the prediction of evaporation from the dam reservoir, and there are several developed models to predict evaporation under different scenarios. Using the deep neural network model long short term memory (LSTM) with a series of modern meta-heuristic algorithms, the present research aims to achieve an acceptable accuracy on weather parameters. The present study investigates the ability of the LSTM deep neural network and meta-heuristic algorithms of Artificial Bee Colony (ABC), Horse Herd Optimization (HOA), and Marine Predators (MPA) to develop an evaporation rate forecasting model for a tropical area in the Dez Dam reservoir, Khuzestan province, Iran. To investigate the effect of different input variable patterns on the prediction accuracy of the case model, twenty-eight scenarios were investigated for the input architecture along with the LSTM model and its combination with meta-heuristic algorithms. The models include ABC-LSTM, LSTM-HOA, and LSTM-MPA, respectively. The results showed that the combined LSTM-MPA model with RMSE, MAE, KGE, and WI evaluation criteria equal to 81.926 mm/month, 53.684 mm/month, 0.850, and 0.816, respectively, performs better than other meta-heuristic models. The examined deep meta-heuristic models are suggested for predicting evaporation from the Dez dam reservoir with limited information because the results show that they have more appropriate prediction capability than the LSTM Model. [ABSTRACT FROM AUTHOR]
- Published
- 2023
- Full Text
- View/download PDF