River Flow Forecasting Using the Gated Recurrent Unit Model with Hybrid Particle Swarm Optimization: The Case Study of Ceyhan Basin

Abstract

One of the most important methods of efficient use of water resources is the effective implementation of watershed-based management. The sustainability of water resources reveals the importance of stream flow estimations. In this study, a hybrid model was proposed to river flow estimation. Deep learning methods named, gated recurrent unit (GRU) and particle swarm algorithm (PSO), are hybridized. In the study, daily flow data of the Fırnız River and Aksu River, flow measurement stations, which are located on different branches of the Ceyhan Basin, were used with the timespan of 2001-2010. Benchmark model (GRU) was compared with hybrid model (PSO-GRU) and linear regression (LR) which is one of the classical methods. Once the results were compared, it was observed that the hybrid model was more successful than the comparison and linear regression models. In addition, the hybrid model confirmed this success according to the RMSE, MAE, MAPE, SD and R2 values, which are among the evaluation criteria.