Chuan Ding, Yi Zhang, Beini Lu, Yijing Feng, Wenhan Li, Jianfeng Peng, Hanlin Huang, Zinuo Cheng, Lin Li, Yeqing Li*, Lu Feng, Hongjun Zhou* and Chunming Xu,
{"title":"AI Data-Driven Based In-Depth Interpretation and Inverse Design for Hydrogen Yield from Biogas Direct Reforming","authors":"Chuan Ding, Yi Zhang, Beini Lu, Yijing Feng, Wenhan Li, Jianfeng Peng, Hanlin Huang, Zinuo Cheng, Lin Li, Yeqing Li*, Lu Feng, Hongjun Zhou* and Chunming Xu, ","doi":"10.1021/acssusresmgt.4c0022710.1021/acssusresmgt.4c00227","DOIUrl":null,"url":null,"abstract":"<p >Exploring the intricate mechanism of factors coordinating with each other and optimizing the reaction conditions are critical to improving the performance of the hydrogen yield from biogas direct reforming (HY-B). Due to the lack of mature direct biogas hydrogen production engineering cases in China, the study data were obtained from a self-constructed HY-B unit that lasted for 42 days with a total of 298 data. In this study, an automated machine learning algorithm (AutoGluon) was used to comprehensively predict and analyze the parameters of HY-B. The study found that the optimal first-layer model is neural network (NN), and the optimal second-layer model is WeightedEnsemble. Meanwhile, based on the Shapley additive explanations (SHAP) values, it was demonstrated that the optimal parameter combination was a temperature range of 900–950 °C, pressure range of 0.15–0.3 bar, and water flow rate of around 24 g/h, which could give a distinguished conversion rate of CH<sub>4</sub> and hydrogen yield. In addition, the experimental verification showed that the Hydrogen-Seek strategy based on the multiobjective particle swarm optimization (MOPSO) could accurately excavate the best process parameters and optimize the combination of conditions and then result in significant improvements. The optimized data set can improve the yield from 63.45% to 67.69%, compared to the highest hydrogen yield in the previous experiment. Our results show that artificial intelligence algorithms can be successfully implemented to predict and improve HY-B performance, and hopefully provide guidance for the intelligent operation of industrial processes in the future.</p>","PeriodicalId":100015,"journal":{"name":"ACS Sustainable Resource Management","volume":"1 11","pages":"2384–2393 2384–2393"},"PeriodicalIF":0.0000,"publicationDate":"2024-10-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"ACS Sustainable Resource Management","FirstCategoryId":"1085","ListUrlMain":"https://pubs.acs.org/doi/10.1021/acssusresmgt.4c00227","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
Exploring the intricate mechanism of factors coordinating with each other and optimizing the reaction conditions are critical to improving the performance of the hydrogen yield from biogas direct reforming (HY-B). Due to the lack of mature direct biogas hydrogen production engineering cases in China, the study data were obtained from a self-constructed HY-B unit that lasted for 42 days with a total of 298 data. In this study, an automated machine learning algorithm (AutoGluon) was used to comprehensively predict and analyze the parameters of HY-B. The study found that the optimal first-layer model is neural network (NN), and the optimal second-layer model is WeightedEnsemble. Meanwhile, based on the Shapley additive explanations (SHAP) values, it was demonstrated that the optimal parameter combination was a temperature range of 900–950 °C, pressure range of 0.15–0.3 bar, and water flow rate of around 24 g/h, which could give a distinguished conversion rate of CH4 and hydrogen yield. In addition, the experimental verification showed that the Hydrogen-Seek strategy based on the multiobjective particle swarm optimization (MOPSO) could accurately excavate the best process parameters and optimize the combination of conditions and then result in significant improvements. The optimized data set can improve the yield from 63.45% to 67.69%, compared to the highest hydrogen yield in the previous experiment. Our results show that artificial intelligence algorithms can be successfully implemented to predict and improve HY-B performance, and hopefully provide guidance for the intelligent operation of industrial processes in the future.