Optimal design of large-scale solar-aided hydrogen production process via machine learning based optimisation framework. (1st January 2022)

• Record Type: Journal Article
• Authors: Wang, Wanrong; Ma, Yingjie; Maroufmashat, Azadeh; Zhang, Nan; Li, Jie; Xiao, Xin
• Is Part Of: Applied energy. Volume 305(2022)
• Publication Date: 2022-01-01
• Subjects: Solar energy -- Hydrogen -- Machine learning -- Hybrid optimization algorithm -- Surrogate model; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2021.117751
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: Four solar steam methane reforming alternatives are investigated. Machine learning based optimisation framework is proposed to achieve optimal design. Total annualised cost is reduced by 14.9 % ~ 15.1% in comparison to existing work. CO2 emission decreases by 80.0 kt yr −1 than conventional steam methane reforming. Levelized cost of H2 production using solar is reduced from 2.9 to 2.4 $ kg −1. Abstract: Hydrogen is an important energy carrier in the transportation sector and an essential industrial feedstock for petroleum refineries, methanol, and ammonia production. Renewable energy sources, especially solar energy have been investigated for large-scale hydrogen production in thermochemical, electrochemical, or photochemical manners due to considerable greenhouse gas emissions from the conventional steam reforming of natural gas and oil-based feedstock. The solar steam methane reforming using molten salt (SSMR-MS) is superior due to its unlimited operation hours and lower total annualized cost (TAC). In this work, we extend the existing optimisation framework for optimal design of SSMR-MS in which machine learning techniques are employed to describe the relationship between solar-related cost and molten salt heat duty and establish relationships of TAC, hydrogen production rate and molten salt heat duty with independent input variables in the whole flowsheet based on 18, 619 sample points generated using the Latin hypercube sampling technique. A hybrid global optimisation algorithm is adopted to optimise the developed model and generate the optimal design, which is validated in SAM and Aspen Plus V8.8. The computational results demonstrate that a significant reduction in TAC by 14.9 % ~ 15.1 %, and CO2 emissions by 4.4 % ~ 5.2 % can be achieved compared to the existing SSMR-MS. The lowest Levelized cost of Hydrogen Production is 2.4 $ kg −1 which is reduced by around 17.2 % compared to the existing process with levelized cost of 2.9 $ kg −1 .