基于机器学习的汽油加氢裂化辛烷值损失预测和脱硫优化  

Prediction of Octane Loss and Optimization of Desulfurization in Gasoline Hydrocracking Based on Machine Learning

作　　者：龙梦舒 闵超[1,2] 赵伟 张馨慧 代博仁 LONG Meng-shu;MIN Chao;ZHAO Wei;ZHANG Xin-hui;DAI Bo-ren(School of Science, Southwest Petroleum University, Chengdu 610000, China;Artificial Intelligence Research Institute, Southwest Petroleum University, Chengdu 610500, China;Shengli Oilfield Exploration and Development Research Institute, Dongying 257000, China)

机构地区：[1]西南石油大学理学院,成都610000 [2]西南石油大学人工智能研究院,成都610500 [3]胜利油田勘探开发研究院,东营257000

出　　处：《科学技术与工程》2022年第3期1076-1084,共9页Science Technology and Engineering

基　　金：成都市国际合作项目(2020-GH02-00023-HZ)。

摘　　要：辛烷值损失的准确预测有助于汽油炼制过程的优化与控制,以达到更好的脱硫效果。原油的加氢脱硫是一个十分复杂的物化反应过程,对于该过程中的参数控制多依赖于工人的经验,因此基于大数据建立辛烷值损失预测模型可以用于优化脱硫效果,从而提高产品质量,减轻工人的劳动强度,具有十分重大的实际意义。采用单因素分析、方差过滤、随机森林等方法进行了特征筛选,最后基于逻辑回归、BP(back propagation)神经网络以及支持向量机(support vector machine,SVM)三种机器学习算法构建了辛烷值损失预测模型。实验结果表明,基于SVM建立的辛烷值损失预测模型精度达到了98.24%,优于逻辑回归和BP神经网络预测模型。将该模型应用于脱硫优化,在生成汽油的硫含量达标的情况下,获得最优的控制变量组合,达到将辛烷值损失降到最低的目的。In order to help optimize and control the gasoline refining process to achieve a better desulfurization effect,the octane loss must be accurately predicted.The hydrodesulfurization of crude oil is a very complex physical and chemical reaction process,and the parameter control in the process mostly depends on the experience of the workers.Therefore,the establishment of an octane loss prediction model based on big data can be used to optimize the desulfurization effect,thereby product quality can be improved and the labor intensity of workers can be reduced,which is of great practical significance.Feature selection was carried out using methods such as single factor analysis,variance filtering,random forest and others,and finally the octane loss prediction model was established based on three machine learning algorithms including logistic regression,BP neural network and support vector machine(SVM).Experimental results show that the prediction accuracy of octane loss based on SVM reaches 98.24%,which is better than logistic regression and BP neural network prediction models.The model is applied to desulfurization optimization,and when the sulfur content of the generated gasoline meets the standard,the optimal control variable combination is obtained to achieve the goal of the lowest octane loss.

关 键 词：辛烷值 预测 加氢脱硫 机器学习 优化 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]