基于BP神经网络的汽油辛烷值损失预测模型研究

Research on Forecast Model of Gasoline Octane Loss Based on BP Neural Network

针对第十七届中国研究生数学建模竞赛降低汽油精制过程中的辛烷值损失问题的研究,基于某企业催化裂化汽油精制过程提供的325个样本的原料油、吸附剂性质和装置各位点实时操作数据,首先利用数据挖掘技术的嵌入式特征选择法、Pearson相关系数分析法降维筛选出14个变量,以此建立辛烷值损失的BP神经网络预测模型.接着用遗传算法模型对影响辛烷值损失值和硫含量的主要变量的操作条件进行优化,使得在辛烷值损失最小的情况下,获得最优催化裂化汽油的操作条件.

The authors investigated the research on the problem of reducing gasoline octane in the gasoline refining process of the 17th China Postgraduate Mathematical Modeling Contest.This paper is based on the 325 samples of raw oil,adsorbent properties and equipment provided by a company's catalytic cracking gasoline refining process.First,14 variables are selected by using the embedded feature selection method of data mining technology and the Pearson correlation coefficient method to reduce the dimensionality to establish a BP neural network prediction model.Then,the genetic algorithm model is used to optimize the operating conditions of the main variables that affect the octane loss and the sulfur content,so that the optimal FCC gasoline operating conditions can be obtained with the minimum octane loss.