Analyzing the Levelized Cost of Hydrogen: Leveraging Machine Learning for Evaluating Production Pathways

Abstract This study examines the economic viability of producing hydrogen by various routes, paying attention to the hydrogen colors that are less prominent in literature. The analysis was done by comparing the levelized cost of hydrogen (LCOH) for eleven hydrogen production pathways. Predictive analysis for hydrogen production rates was also carried out using an artificial neural network (ANN) model. The data was collected using a literature-based methodology which considered several factors, such as feedstock costs, operating expenses, and capital expenditures. The study showed that the LCOH varies greatly depending on the production pathways. Due to its more modern technology, hydrogen obtained from fossil fuels (black/brown, gray, and blue) is usually less expensive than hydrogen derived from renewable energy sources (green, yellow). On the other hand, blue hydrogen requires carbon capture technology, which makes the operating costs associated with it higher. Aqua hydrogen stands out with the lowest levelized cost of $0.23/kg hydrogen produced, but environmental concerns from oil sand extraction and emerging technologies like orange hydrogen need further consideration. The ANN model was developed based on key hydrogen production parameters, and its performance was evaluated using metrics such as the proportion of variance (R-squared), mean square error and root mean square error. The R-squared result for the test data set was 92% (R2=0.92), showing the model's high predictive accuracy. This study is unique in two ways: first, it shows significant differences in LCOH across several hydrogen colors, with aqua hydrogen becoming the most economical choice. Second, it shows how machine learning can be used to find cost-cutting measures and optimize hydrogen production processes through various methods. Further research is required to address the lack of real-time production data, investigate the environmental effects of these production processes, and improve the accuracy of developed models.