A comprehensive model integrating BP neural network and RSM for the prediction and optimization of syngas quality

Abstract

A comprehensive model for the prediction and optimization of high-quality syngas is developed from feedstock selection to operating conditions selection. This is a study to explore the integration of a back propagation (BP) neural network and response surface methodology (RSM) in biomass gasification. In this paper, the input variables, namely, exergy efficiency of hydrogen and cold gas efficiency (CGE) of combustible syngas, were used as the evaluation indexes of syngas quality. The datasets for the study of the prediction and optimization were generated by running a validated Aspen Plus process model. First, 110 biomasses with different compositions were simulated in the process model under specific conditions. Then, the correlation between the input and output variables was analysed and verified through partial correlation analysis and linear fitting. For the prediction of syngas quality, taking the content of C, H, and O as input variables, 110 sets of data were applied in the BP neural network. Regarding the optimization of syngas quality, an RSM was introduced with the gasification temperature, steam-to-biomass (S/B) ratio, and equivalence ratio (ER) as input variables. The results of the BP neural network showed that it can realize the prediction function of syngas quality with good accuracy (R > 0.98 and R > 0.99, respectively). The results of the RSM indicated that the optimum conditions for high-quality syngas were T = 822 °C, S/B = 0.2, ER = 0.01 and T = 885 °C, S/B = 0.2, ER = 0.01.