Proposed CGU (CNN-GRU Unified) Model with NAS-Based Bayesian Optimization

Published in Egyptian Informatics Journal: A novel unified deep learning model (CGU) optimized via NAS-based Bayesian techniques for hydrological forecasting.

This article, published in the Egyptian Informatics Journal (Elsevier), proposes an original CGU (CNN-GRU Unified) model developed by me to address the shortcomings of existing hybrid models in the literature. The study distinguishes itself by employing a NAS (Neural Architecture Search) based Bayesian optimization approach that automates not only the hyperparameters but also the architectural structure of the model. Developed Original Architecture and Method: Proposed Model (CGU): An integrated (Unified) architecture was designed that combines the spatial feature extraction of CNN layers with the temporal memory capability of GRU, operating more efficiently than standard CNN-LSTM structures. NAS-Based Optimization: Architectural components such as the number of layers, number of neurons, and activation functions of the model were determined not through manual trials, but through Bayesian optimization using the Neural Architecture Search algorithm. Outstanding Performance: The proposed CGU model was tested on the Yusufeli and Deriner dams; Compared to classical LSTM and CNN models, it achieved an accuracy score of over 99.41% by minimizing prediction errors.