Autoencoder based cognitive deep learning approach for estimating hydrogen-enriched natural gas from thermal energy sources

Geothermal resources can be used to produce the hydrogen needed for hydrogen-enriched natural gas (HENG). In this study, three models with a cognitive deep learning (CDL) approach were developed to estimate the amount of HENG that can be obtained based on geothermal energy-assisted hydrogen production. The main data was recorded from a currently working plant. CDL is a technique for incorporating human-like cognitive skills into artificial intelligence models. CDL combines deep learning models with cognitive concepts to improve pattern identification, prediction, and decision-making processes, particularly in large-scale data processing. Autoencoders (AEs), deep belief networks (DBNs), feedforward neural networks (FFNNs), gated recurrent units, and echo state networks were used for developing models with a CDL approach. In these models, AEs were used to compress and reconstruct multidimensional data in hidden space, achieving a nonlinear dimensionality reduction process. The approach utilized in this study aimed to provide a scalable and accurate tool for renewable hydrogen production based on HENG estimation, considering the growing global energy demand. As a result of the analysis, AE-DBN-FFNN model achieved a significant success rate in determining the amount of HENG by exhibiting a superior estimation performance with an R2 value of 0.9887 and RMSE value of 18.9166.