Forecasting Dam Storage Volume Using a Hybrid RNN Model Empowered by Tunable Q-Factor Wavelet Transform and Metaheuristic Optimization

It is universally acknowledged that water is essential for the survival of humanity. Therefore, the effective utilization and sustainability of water resources are of paramount importance. Driven by this necessity, this study develops predictive models for the Çubuk2 Dam, which supplies drinking water to Ankara, using historical data on temperature, precipitation, humidity, reservoir volume, and water level. The models were constructed using deep learning, optimization, and wavelet decomposition techniques, which have garnered significant attention from researchers in recent years. Specifically, Recurrent Neural Network (RNN), Random Forest (RF), Particle Swarm Optimization (PSO), and Tunable Q-Factor Wavelet Transform (TQW) methods were utilized. RNN was employed both as a standalone model and hybridized as RNNRF and RNNPSO, with TQW applied to all models to enhance predictive performance. Furthermore, ten different input scenario structures were established using Mutual Information (MI). To evaluate model performance, the Correlation Coefficient (R), Nash–Sutcliffe Efficiency (NSE), Kling–Gupta Efficiency (KGE), Performance Index (PI), and Root Mean Square Error (RMSE) metrics were adopted. The results demonstrated that the hybrid models yielded highly effective outcomes and that Mutual Information successfully identified optimal model input structures.