Comparative analysis of machine learning and artificial intelligence models for optimizing mixed refrigerant characteristics in a hydrogen pre-cooling storage system

Abstract

In the current era of renewable energy transition, efficient hydrogen storage is indispensable for overcoming intermittent challenges and enabling the widespread adoption of clean energy technologies. Machine learning is crucial for optimizing hydrogen storage by providing predictive analytics and data-driven insights to enhance storage efficiency and address complex operational challenges. This study would evaluate various machine learning and artificial intelligence techniques, such as neural networks, genetic algorithms, Support Vector Machines, and Response Surface Methodology to optimize the composition of mixed refrigerants for cryocompressed hydrogen storage. The research would aim to determine which models demonstrate superior performance in predicting the most effective refrigerant compositions for maximizing storage capacity, efficiency, and safety while minimizing energy consumption and environmental impact. The Method based on Removal Effect of Criteria (MEREC) method will be employed to ascertain priority weights, which will then be inputted into K-means machine learning clustering algorithms for model grouping, optimizing efficiency and organization in hydrogen storage research and development. The optimization results based on the objective function developed indicate that a feasible combination approximately is 30% C2*, 20% C1, 12 % C3, 10% n-C4, 8 % iC5, 13 % N2, 5 % R14, and 2 % H2. The MEREC method assigned weightage percentages to form clusters with relevance holding the highest priority at 0.37, followed by skill requirement at 0.23. Sensitivity analysis was performed to examine the prioritization of the best models identified from the cluster analysis. From the best cluster i.e. cluster one, among the five best models SVM- model 7 showed the best result. The concluding remarks deduced that SVM is the best model and it was followed by adaptive network-based fuzzy inference systemGenetic algorithm (ANFIS-GA). By comparing the efficacy of different ML and AI approaches, the study could provide valuable insights into the most suitable methodologies for designing efficient and reliable cryocompressed hydrogen storage systems.