基于数据驱动的乙二醇精馏过程能耗与产品质量建模  

Data driven modeling of energy consumption and product quality in ethylene glycol distillation process

作　　者：冯康康 耿欣 娄清辉 王玉 胡华军 石祥建 薄翠梅 Kangkang FENG;Xin GENG;Qinghui LOU;Yu WANG;Huajun HU;Xiangjian SHI;Cuimei BO(Nanjing Nanrui Jibao Engineering Technology Co.,Ltd,Nanjing,Jiangsu 211102,China;College of Electrical Engineering and Control Science,Nanjing Tech University,Nanjing,Jiangsu 211800,China)

机构地区：[1]南京南瑞继保工程技术有限公司,江苏南京211102 [2]南京工业大学电气工程与控制科学学院,江苏南京211800

出　　处：《过程工程学报》2025年第2期142-149,共8页The Chinese Journal of Process Engineering

摘　　要：能耗和产品质量是精馏过程中的重要经济指标,能耗和产品质量预测是实现工艺优化操作的必要环节。以煤制乙二醇精馏过程为例,采用一种改进的最小二乘支持向量机算法,构建乙二醇精馏过程能耗与产品质量模型。首先基于煤制乙二醇精馏过程的工业数据,采用互信息法提取关键特征参数,进行变量筛选和数据预处理。然后引入局部目标集提高模型预测精度,并通过UMDA算法进行迭代寻优,得到LSSVM模型的最佳超参数。最后采用改进的LSSVM算法对数据样本建模,验证了算法的有效性。后续可以将模型作为多目标优化问题的目标函数求解最佳操作参数,来达到稳定操作、提高产品质量、避免过度分离操作以及降低精馏过程能耗的目的。With the rapid development of polyester industry,the increasing demand of ethylene glycol(EG)is in conflict with the shortage of supply in China.Large project of EG production from coal has been receiving more and more attention.In the production of coal-to-ethylene glycol,the optimisation of distillation operations represents a vital means to achieve energy saving and consumption reduction,as well as quality enhancement and efficiency improvement.The foundation of optimisation lies in the establishment of precise models for the process.However,due to the complex reactions,strong system coupling,and non-linearity inherent in the distillation process,it is difficult to accurately construct models using traditional mechanistic methods.Therefore,this study uses the distillation process of coal-to-ethylene glycol as the research subject,employing a refined Least Squares Support Vector Machine(LSSVM)algorithm to accurately construct energy consumption and product quality models for the ethylene glycol distillation process.In this process,the actual industrial data from the coal-to-ethylene glycol distillation process was used as the benchmark,the mutual information method was employed to extract the main feature parameters,and variable screening and data pre-processing were conducted.Subsequently,by introducing local target sets and using the UMDA algorithm for iterative optimisation,the optimal hyperparameters were derived.After determining the optimal hyperparameters,the improved LSSVM algorithm was used to model the data samples and further compared this model with other purity and energy consumption models established by different algorithms.This comparison confirmed the high efficiency and accuracy of the improved LSSVM algorithm based on UMDA proposed in this paper.In summary,compared with traditional Support Vector Machine methods,the LOS-LSSVM model based on the UMDA optimisation process has a clear advantage in data fitting,accurately reflecting the actual situation of the distillation process and effectively

关 键 词：煤制乙二醇 数据驱动建模 最小二乘支持向量机 能耗 质量 

分 类 号：TQ223.162[化学工程—有机化工] TQ028.31[自动化与计算机技术—控制理论与控制工程] TP181[自动化与计算机技术—控制科学与工程]