Enhancing Streamflow Prediction Accuracy: A Comprehensive Analysis of Hybrid Neural Network Models with Runge–Kutta with Aquila Optimizer

Abstract

Abstract This study investigates the efficacy of hybrid artificial neural network (ANN) methods, incorporating metaheuristic algorithms such as particle swarm optimization (PSO), genetic algorithm (GA), gray wolf optimizer (GWO), Aquila optimizer (AO), Runge–Kutta (RUN), and the novel ANN-based Runge–Kutta with Aquila optimizer (LSTM-RUNAO). The key novelty of this research lies in the developing and applying the LSTM-RUNAO model, which combines Runge–Kutta and Aquila optimizer to enhance streamflow prediction accuracy. The models’ performance is compared against the conventional ANN method, analyzing monthly streamflow prediction across three data split scenarios (50–50%, 60–40%, and 75–25%). Results show that the LSTM-RUNAO model outperformed conventional ANN methods, achieving a 28.7% reduction in root mean square error (RMSE) and a 20.3% reduction in mean absolute error (MAE) compared to standard ANN models. In addition, the model yielded a Nash–Sutcliffe Efficiency (NSE) improvement of 12.4% and an R-squared value increase of 7.8%. The study advocates for the 75–25% train-test data splitting scenario for optimal performance in data-driven methodologies. Furthermore, it elucidates the nuanced influence of input variables on prediction accuracy, emphasizing the importance of thoughtful consideration during model development. In summary, this research contributes valuable insights and introduces an innovative hybrid model to enhance the reliability of streamflow prediction models for practical applications.

Keywords

• Streamflow prediction
• Neural networks
• Data splitting
• Runge–Kutta with Aquila optimizer