Streamflow forecasting using a hybrid LSTM-PSO approach: the case of Seyhan Basin

The conditions which affect the sustainability of water cause a number of serious environmental and hydrological problems. Effective and correct management of water resources constitutes an effective and important issue among scales. In this sense, a precise estimation of streamflow time series in rivers is one of the most important issues in optimal management of surface water resources. Therefore, a hybrid method combining particle swarm algorithm (PSO) and long short-term memory networks (LSTM) are proposed to predict flow with data obtained from different flow measurement stations. In this respect, the data gathered from three Flow Measurement Stations (FMS) from Zamanti and Eglence rivers located on Seyhan Basin are utilized. Besides, the proposed LSTM-PSO method is compared to an adaptive neuro-fuzzy inference system (ANFIS) and the LSTM benchmark model to demonstrate the performance achievement of proposed method. The prediction performances of the developed hybrid model and the others are tested on the determined stations. The forecasting performances of the models are determined with RMSE, MAE, MAPE, SD, and R-2 metrics. The comparison results indicated that the LSTM-PSO method provides highest results with values of R-2 (approximate to 0.9433), R-2 (approximate to 0.6972), and R-2 (approximate to 0.9273) for the Degirmenocagi, Egribuk, and Ergenusagi FMS data, respectively.