Optimal Model of C4 Olefins based on Neural Network Pick, Want 

Xuejian Gou*, Mingyue Yang, Senlin Deng 

Lanzhou University of Technology, Lanzhou, 730050, China 

Abstract: In the field of chemical industry and medicine, C4 olefin has high utilization value. In the prepara-tion of C4 olefin, different catalyst and temperature combinations have a significant impact on the yield of C4 olefin. Therefore, it is of important practical significance to choose reasonable catalyst and temperature com-binations in production. For problem 1, the neural network is used to machine learn the training set separated from the processed data, and find the sum of squares of the difference between the predicted value and the ac-tual value of each model by least squares method, and obtain the optimal prediction model after comparison. Then the five parameter combination input model with a certain step length was predicted to obtain all the dif-ferent temperature and catalyst combinations to obtain the corresponding C4 olefin yield.By comparison, when the catalyst combination was 150mg 3wt%Co / SiO2-170mg HAP-ethanol concentration of 0.3 ml/min and a temperature of 370 degrees, the C4 olefin yield reached a maximum of 0.99976. When the temperature was controlled below 350 degrees, the results were continuously optimized by shortening the step length of the temperature, and finally obtained the C4 olefin yield reached a maximum of 0.27202 when the catalyst was 190mg 3wt%Co / SiO2-190mg HAP-ethanol concentration of 1.9 ml/min and a temperature of 349 de-grees. For problem 2, the two experiment were designed to verify the optimal model and the C4 olefin yield as high as possible. Experiments 1,2 were designed to verify whether the catalyst combination with the peak C4 olefin yield, predicted by Problem 3 models, was close to the actual C4 olefin yield. Second, to compen-sate for the variables that cannot be explored in Annex 1, we designed three experiments based on the data in Annex 1. Finally, this paper explores the sensitivity of the model to improve the future by exploring the ad-vantages and disadvantages of the four independent variables in the catalyst. 

Keywords: Interpolation fit; Multiple linear regression analysis; Neural network; Combinatorial optimization