Application of multiple regression modeling and BP neural network in the analysis of prepared C4 olefins

Abstract Ethanol serves as a crucial raw material for synthesizing C4 olefins, making the investigation of catalytic synthesis process conditions highly valuable. This study conducted a correlation analysis on experimental data from the catalytic coupling of ethanol into C4 olefins under varying temperatures and catalyst conditions, determining the Pearson correlation coefficient between ethanol conversion, C4 olefin selectivity, and temperature. The analysis revealed a significant relationship between C4 olefin selectivity and temperature. Following this, gray correlation analysis was employed to pinpoint the experimental conditions with the highest correlation coefficients. Using six independent variables, including catalyst combinations and temperature, alongside dependent variables such as ethanol conversion rate and C4 olefin selectivity, a multiple regression model was established. Results indicated that temperature significantly influences the dependent variables, with Co/SiO 2 dosage and Co loading being the most pivotal factors affecting ethanol conversion. Specifically, while Co/SiO 2 and Co loading have minimal impact on ethanol conversion, HAP dosage and temperature are positively correlated. Ethanol dosage is negatively correlated. Additionally, C4 olefin selectivity is less influenced by the amounts of Co/SiO 2 and ethanol. Finally, a neural network model was developed linking five influencing factors to C4 olefin yield. Bayesian regularization was used to train the model, optimizing the experimental conditions for maximizing C4 olefin yield.