Optimization problem of C4 olefin production based on BP neural network prediction model

Abstract

This paper focuses on the effects of different temperature and catalyst combinations on ethanol conversion, C4 olefin selectivity and C4 olefin yield. Theoretical references are provided for practical production and experiments. In order to study the relationship between ethanol conversion and C4 olefin selectivity under each catalyst combination condition and temperature more intuitively, a BP neural network prediction model was established to pre-process the experimental data, and it can be analyzed from the function plot that both show positive correlation with temperature under the same catalyst combination. On this basis, the experimental results provided by experiment two were briefly analyzed. On the basis of this model, the functional steepness on different functional segments represented by different catalyst combinations were compared, and the steeper the functional segment, the more effective the catalyst combination. The analysis showed that when the loading ratio of Co/SiO2 to HAP was close to 1:1 and the Co loading and ethanol concentration were close to 1.68 ml/min, it was more favorable to improve the ethanol conversion and C4 olefin selectivity. In analyzing the effect of different temperatures on the reaction, all the predicted data images were superimposed and it was found that the higher the temperature, the better the ethanol conversion and C4 olefin selectivity. Using a genetic algorithm model, the best solution was determined: a charge ratio of 1.1403 for Co/SiO2 and HAP, a Co loading of 1.9276, an ethanol concentration of 1.9472, and a temperature of 322.21 degrees C. The yield of C4 olefins obtained was about 53.04%. In order to achieve the goal of using less raw material in production and ultimately maximizing profits, the search for the optimal solution continued based on the previous study. A neural network prediction model was used to find and then breakthrough around this small region, and five sets of experiments were designed.