Determination of hydrogen production in power plant using predictive machine learning methods

Abstract With increasing global warming, the focus on sustainable and eco-friendly energy solutions is raised. The production of hydrogen is identified to be extremely important due to its role in efforts to reduce greenhouse gas emissions, hence mitigating climate change. Predicting the amount of hydrogen production accurately in this context is essential. This requires developing and/or applying complex predictive tactics to improve hydrogen production related to reliability and efficiency in thermal power plants. Hydrogen production quantification in thermal power plants using advanced predictive machine learning (ML) tactics is the focus of this exploration. The CatBoost, Random Forest (RF), Long Short-Term Memory (LSTM), and Bootstrap aggregating tactics are used to project hydrogen production. On the other hand, the Firefly algorithm is utilized to fine-tune the hyperparameters of these schemes. Predictions are made on a simulated dataset collected from the recommended power plant, and the accuracy of the schemes’ predictions is assessed using different evaluation metrics. Regarding the results, it turns out that the Firefly-CB model performs very well, having near-perfect R 2 value of 0.9764, proving the capability of predicting hydrogen with small errors. Nevertheless, the Firefly-LSTM model performs poorly with the lowest R 2 = 0.5221 and highest errors during data test. The productivity of Firefly-CB is also compared with that of other advanced optimizers and ensemble schemes, in which Firefly-CB again reveals its strength in the recommended cycle’s hydrogen prediction. These findings from the study demonstrate that hybrid ML schemes can significantly improve the prediction of hydrogen production at power stations, contributing to efficient and sustainable energy management.