Application of machine learning in evaluating and optimizing the hydrogen production performance of a solar-based electrolyzer system

Abstract

A green hydrogen production method based on solar energy is proposed, and Machine Learning (ML) models are adopted for system optimization. This study assayed to highlight the potential of using ML in anticipating the performance of a Photovoltaic Thermal (PVT) system integrated with an electrolyzer device. For this purpose, the performance of four different ML models, including Extreme Learning Machine (ELM), Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU), and CatBoost, in predicting the hydrogen production of the solar-based system are analyzed. The hyperparameters of the ML models are optimized using the Arithmetic Optimization Algorithm (AOA). Furthermore, the optimum performance of the system is obtained by implementing the Multi-Objective Grey Wolf Optimizer (MOGWO). The outcomes reveal that the AOA-CatBoost is the most accurate ML model in predicting the performance of the system. This model forecasts that the maximum hydrogen production rates of the system in the seasons of spring, summer, fall, and winter are respectively 2.10 mol/h, 2.22 mol/h, 1.93 mol/h, and 1.87 mol/h. Besides, the MOGWO shows that the maximum amount of hydrogen production and electrical generation of the system are 24.48 mol/day and 7.74 MJ/day, respectively.