化工学报 ›› 2019, Vol. 70 ›› Issue (11): 4315-4324.DOI: 10.11949/0438-1157.20190268

• 过程系统工程 • 上一篇    下一篇

模糊神经网络-遗传算法优化丙烯酸苄酯合成工艺

范峥1(),姬盼盼1,李超2,刘壮3,赵毅刚3,井晓燕1   

  1. 1. 西安石油大学化学化工学院,陕西 西安 710065
    2. 中国石油长庆油田分公司气田开发事业部,陕西 西安 710021
    3. 中国石油长庆油田分公司第一采气厂,陕西 榆林 718500
  • 收稿日期:2019-03-26 修回日期:2019-07-20 出版日期:2019-11-05 发布日期:2019-11-05
  • 通讯作者: 范峥
  • 作者简介:范峥(1982—),男,博士,副教授,fanzheng@xsyu.edu.cn
  • 基金资助:
    陕西省科学技术研究与发展计划项目(2016GY-150);西安石油大学研究生创新与实践能力培养项目(YCS18221011);中国国家留学基金项目(201908610135)

Synthetic process optimization of benzyl acrylate using fuzzy neural networks-genetic algorithms

Zheng FAN1(),Panpan JI1,Chao LI2,Zhuang LIU3,Yigang ZHAO3,Xiaoyan JING1   

  1. 1. College of Chemistry & Chemical Engineering, Xi’an Shiyou University, Xi’an 710065, Shaanxi, China
    2. Gas Field Development Office, Changqing Oilfield Company of PetroChina, Xi’an 710021, Shaanxi, China
    3. The First Gas Plant, Changqing Oilfield Company of PetroChina, Yulin 718500, Shaanxi, China
  • Received:2019-03-26 Revised:2019-07-20 Online:2019-11-05 Published:2019-11-05
  • Contact: Zheng FAN

摘要:

首先通过多因素方差分析探讨携水剂用量、反应温度、反应真空度、反应时间、酸醇比对丙烯酸苄酯质量分数及收率的影响,然后以显著因素为输入、综合得分为输出建立Takagi-Sugeno型模糊人工神经网络,最后利用遗传算法优化丙烯酸苄酯合成工艺条件并使用t检验法验证可靠性。研究表明,上述各因素对丙烯酸苄酯合成产物的质量分数与收率同时具有非常显著的影响,预测模型采用5-15-243-1型网络结构,经36859次训练其均方差小于允许收敛误差限0.0050,输出值与期望值呈近似线性关系,训练、测试阶段决定系数0.9999、0.9998。借助遗传算法经149次进化得到最优控制参数,即当携水剂用量为53 ml,反应温度为125℃,反应真空度为0.095 MPa,反应时间为2.2 h,酸醇比为1.4时,丙烯酸苄酯的质量分数、收率及综合得分为99.27%、98.04%、98.78%,经验证该模型亦可靠性良好。

关键词: 丙烯酸苄酯, 合成, 神经网络, 遗传算法, 优化

Abstract:

Firstly, the effects of water-carrying agent dosage, reaction temperature, reaction vacuum degree, reaction time, acid-alcohol ratio on the mass fraction and yield of benzyl acrylate were investigated by multivariate analysis of variance. Then the Takagi-Sugeno fuzzy artificial neural networks was established with significant factors as input and comprehensive scores as output. Finally, the genetic algorithm was used to optimize the synthesis conditions of benzyl acrylate and the reliability was verified by t-test. The research demonstrated that all mentioned factors behaved extremely significant effects on the mass fraction and yield of benzyl acrylate synthesized products simultaneously. The prediction model was based on 5-15-243-1 network structure. After 36859 training iterations, the mean square error of the prediction model was less than the allowable convergence error limit 0.0050. The relationship between the output value and the expected value was approximately linear. The determination coefficient of training and testing stages were 0.9999 and 0.9998, respectively. The optimal control parameters including 53 ml water-carrying agent, 125℃ reaction temperature, 0.095 MPa reaction vacuum degree, 2.2 h reaction time and 1.4 acid-alcohol ratio were obtained by 149 evolutions of genetic algorithm. The mass fraction, yield and comprehensive scores of benzyl acrylate synthesis were 99.27%, 98.04% and 98.78% on the basis of the optimal process conditions. The prediction model was also proved to be reliable.

Key words: benzyl acrylate, synthesis, neural networks, genetic algorithm, optimization

中图分类号: