Chemical Industry and Engineering Progress ›› 2022, Vol. 41 ›› Issue (9): 4691-4700.DOI: 10.16085/j.issn.1000-6613.2021-2466

• Chemical processes and equipment • Previous Articles     Next Articles

Local modeling and optimization of K-means-PSO-SVR for methanol to aromatics

LIU Penglong1(), XU Xiongfei1, ZHANG Wei1(), XU Xin1, ZHANG Kan2, WANG Junwen1   

  1. 1.College of Chemistry and Chemical Engineering, Taiyuan University of Technology, Taiyuan 30024, Shanxi, China
    2.Institute of Coal Chemistry, Chinese Academy of Sciences, Taiyuan 030001, Shanxi, China
  • Received:2021-12-01 Revised:2022-03-03 Online:2022-09-27 Published:2022-09-25
  • Contact: ZHANG Wei

甲醇制芳烃K-means-PSO-SVR局部建模及优化

刘鹏龙1(), 许雄飞1, 张玮1(), 许鑫1, 张侃2, 王俊文1   

  1. 1.太原理工大学化学化工学院,山西 太原 030024
    2.中国科学院山西煤炭化学研究所,山西 太原 030001
  • 通讯作者: 张玮
  • 作者简介:刘鹏龙(1997—),男,硕士研究生,研究方向为化工过程建模与优化。E-mail:642213917@qq.com
  • 基金资助:
    山西省重点研发计划(201903D121027);中科院关键技术人才项目(YB2021001);化学工程联合国家重点实验室开放课题(SKL-ChE-21A01);国家自然科学基金(22178241)

Abstract:

Aiming at the characteristics of sample data convergence, high dimension, nonlinear, strong coupling and large local difference in methanol to aromatics (MTA) process, we proposed a local modeling method of K-means-PSO-SVR to solve the problems of low prediction accuracy and weak robustness of single global model. Firstly, K-means algorithm was used to cluster the data in the sample space and divide in to k regions. Then, the support vector regression algorithm (SVR) optimized by particle swarm optimization (PSO) was used to establish mutually independent local models in the divided sample space. Finally, the k mutually independent local models were combined to form an integrated model covering the whole sample space. Under different noise levels, the performance of K-means-PSO-SVR method was compared with that of single global SVR, BP neural network and linear regression algorithm. The results showed that the K-means-PSO-SVR local modeling method was significantly better than the other three modeling methods at all levels of noise with strong robustness to noise. On the basis of the established data model, the key process parameters of the two-stage fixed-bed MTA were optimized which were verified by five independent repeated experiments. When the first and second stage reactor temperatures were 446.2℃ and 467.3℃ respectively, the volume space velocity of methanol was 0.4h-1 and the pressure was 0.64MPa, and the highest yield total yield of benzene, toluene and xylene (BTX) was 44.30%.

Key words: methanol to aromatics, local modeling, algorithm, model, optimization

摘要:

针对甲醇制芳烃(MTA)过程数据样本趋同、维度高、非线性、强耦合、局部差异大的特性,提出了一种K-means-PSO-SVR的局部建模方法,用以解决单一全局模型预测精度低,鲁棒性不强的问题。该方法首先用K-means算法对样本空间的数据进行聚类,实现对样本空间k个区域的划分,再用经过粒子群优化算法(PSO)优化过超参数的支持向量回归算法(SVR)在划分好的样本空间上建立相互独立的局部模型,最终将建立的k个相互独立的局部模型组合起来组成覆盖整个样本空间的集成模型。在不同噪声水平下将K-means-PSO-SVR方法的建模效果与单一全局SVR、BP神经网络和线性回归3种算法的建模效果进行了比较分析,结果表明:K-means-PSO-SVR局部建模方法的性能在所有水平的噪声下都明显优于其他3种建模方法,并且该方法对噪声具有很强的鲁棒性。以建立的数据模型为基础优化了两段式固定床甲醇制芳烃的关键工艺参数,并用5次独立重复实验验证了优化结果的可靠性,得出当一段温度为446.2℃、二段温度为467.3℃、甲醇体积空速为0.4h-1、压力为0.64MPa时反应产物中苯、甲苯和二甲苯(BTX)的总收率最高,最高收率为44.30%。

关键词: 甲醇制芳烃, 局部建模, 算法, 模型, 优化

CLC Number: 

京ICP备12046843号-2;京公网安备 11010102001994号
Copyright © Chemical Industry and Engineering Progress, All Rights Reserved.
E-mail: hgjz@cip.com.cn
Powered by Beijing Magtech Co. Ltd