Carbon dioxide-abated hydrogen can be synthesised via various processes,one of which is sorption enhanced steam methane reforming(SE-SMR),which produces separated streams of high purity H_(2) and CO_(2).Properties of ...Carbon dioxide-abated hydrogen can be synthesised via various processes,one of which is sorption enhanced steam methane reforming(SE-SMR),which produces separated streams of high purity H_(2) and CO_(2).Properties of hydrogen and the sorbent material hinder the ability to rapidly upscale SE-SMR,therefore the use of artificial intelligence models is useful in order to assist scale up.Advantages of a data driven soft-sensor model over ther-modynamic simulations,is the ability to obtain real time information dependent on actual process conditions.In this study,two soft sensor models have been developed and used to predict and estimate variables that would otherwise be difficult direct measured.Both artificial neural networks and the random forest models were devel-oped as soft sensor prediction models.They were shown to provide good predictions for gas concentrations in the reformer and regenerator reactors of the SE-SMR process using temperature,pressure,steam to carbon ratio and sorbent to carbon ratio as input process features.Both models were very accurate with high R^(2) values,all above 98%.However,the random forest model was more precise in the predictions,with consistently higher R^(2) values and lower mean absolute error(0.002-0.014)compared to the neural network model(0.005-0.024).展开更多
文摘Carbon dioxide-abated hydrogen can be synthesised via various processes,one of which is sorption enhanced steam methane reforming(SE-SMR),which produces separated streams of high purity H_(2) and CO_(2).Properties of hydrogen and the sorbent material hinder the ability to rapidly upscale SE-SMR,therefore the use of artificial intelligence models is useful in order to assist scale up.Advantages of a data driven soft-sensor model over ther-modynamic simulations,is the ability to obtain real time information dependent on actual process conditions.In this study,two soft sensor models have been developed and used to predict and estimate variables that would otherwise be difficult direct measured.Both artificial neural networks and the random forest models were devel-oped as soft sensor prediction models.They were shown to provide good predictions for gas concentrations in the reformer and regenerator reactors of the SE-SMR process using temperature,pressure,steam to carbon ratio and sorbent to carbon ratio as input process features.Both models were very accurate with high R^(2) values,all above 98%.However,the random forest model was more precise in the predictions,with consistently higher R^(2) values and lower mean absolute error(0.002-0.014)compared to the neural network model(0.005-0.024).
