Prediction Of Catalytic Cracking Gasoline Yield Based On MIPSO-XGBoost Algorithm

Catalytic cracking plays an important role in the production of oil refining industry.Predicting the products of catalytic cracking is conducive to the introduction of better variable combinations,thereby increasing product output and improving the economic benefits of refineries.The traditional prediction model of catalytic cracking mainly focuses on the reaction mechanism of the product.However,for more accurate prediction,it must be divided into a finer lumped number or molecular number which will result in a huge amount of computation.The application of machine learning method to cracking catalysis is able to the analysis of various influencing variables from the view of data.Therefore,this paper applies machine learning method,XGBoost,to predict gasoline yield.First of all,the variables are screened by Pearson correlation analysis,and a total of18 variables including raw materials,temperature,pressure,catalyst,etc.are retained.Secondly,from the perspective of parameter optimization,a mixed integer particle swarm optimization(MIPSO)algorithm with linear decreasing weights is constructed to optimize the parameters in XGBoost.Finally,a gasoline yield prediction model based on MIPSO-XGBoost hybrid algorithm is established.Experimental results are shown below.1.Compared with the classical XGBoost algorithm and PSO-XGBoost algorithm,MIPSO-XGBoost predicts better.In the comparison,it is found that PSO will fall into local optimality in the six-dimensional parameters optimization problem of this paper,and the MIPSO algorithm can escape from the local optima,which also ensures the superiority of the MIPSO algorithm proposed in this paper.2.Compared with the three algorithms of GBDT,Ada Boost,and random forest,MIPSO-XGBoost is 5.6898,5.8295,and 7.0505 smaller in MSE,indicating that MIPSO-XGBoost has obvious advantages in prediction accuracy.3.MIPSO is applied to optimize the parameters of GBDT,Ada Boost,and RF,and then the optimized model is compared with the MIPSO-XGBoost in this paper.The comparison shows that MIPSO-XGBoost has the highest prediction accuracy.Thus,the established MIPSO-XGBoost gasoline production rate prediction model can achieve more accurate and effective prediction.