石油学报(石油加工) ›› 2020, Vol. 36 ›› Issue (4): 878-888.doi: 10.3969/j.issn.1001-8719.2020.04.026

• 综述 • 上一篇    

智能优化算法及人工神经网络在催化裂化模型分析中的应用进展

杨帆1,2,周敏3,金继民2,曹军4   

  1. 1. 四川大学 计算机学院, 四川 成都 610065;
    2. 联想集团数据智能应用实验室,四川 成都 610041;
    3. 四川轻化工大学过程装备与控制 工程四川省朵重点实验室,四川 自贡 643000;
    4. 华东理工大学 机械与动力工程学院,上海 200237
  • 收稿日期:2019-05-17 修回日期:2019-12-12 出版日期:2020-07-25 发布日期:2020-09-23
  • 通讯作者: 曹军,男,助理研究员,博士,从事石油化工及工业大数据相关研究,Tel:021-64253810,E-mail:caojun@ecust.edu.cn E-mail:caojun@ecust.edu.cn
  • 作者简介:第一作者:杨帆,男,博士研究生,从事工业智能相关研究
  • 基金资助:
    上海市自然科学基金项目(18ZR1409000)和过程装备与控制工程四川省高校重点实验室开放基金资助项目(GK201818)资助

Research Progress on Application of Intelligent Optimization Algorithms and Artificial Neural Network in FCC Model Analysis

 YANG Fan1,2, ZHOU Min3, JIN Jimin2, CAO Jun4   

  1.  1. College of Computer Science, Sichuan University, Chengdu 610065;
    2. Data Intelligence Application Lab, Lenovo Group, Chengdu 610041, China;
    3. Sichuan Provincial Key Lab of Process Equipment and Control, Sichuan University of science and Engineering, Zigong 643000, China;
    4. School of Mechanical and Power engineering,East China University of Science and Technology, Shanghai 200237, China
  • Received:2019-05-17 Revised:2019-12-12 Online:2020-07-25 Published:2020-09-23
  • Supported by:
     

摘要: 催化裂化是一个由多种高度非线性和相互强关联因素影响的复杂工艺过程,对其工艺过程和产品收率优化的数学建模分析一直是石油加工领域研究的热点和难点。集总动力学模型是机理分析层面最为常用的研究方法。选用合适而快捷的参数估算和求取方法,是集总动力学模型构建过程中的重要一环。遗传算法、粒子群算法和模拟退火算法等智能算法一定程度上克服了经典算法对初值依赖性,难寻找全局最优的问题,同时还保证了算法的收敛性,对于集总动力学模型的发展起到了极大的促进作用。此外,通过构建原料油性质、催化剂性质、操作条件和产品分布之间的神经网络模型,可以从统计学的角度找到产物分布的影响机制,分析得到常规集总分析方法忽略的一些因素,且可对产物分布进行进一步的预测,是构建催化裂化分析模型的一种新型且有效的手段。笔者对现有关于人工智能算法在催化裂化工艺模型构建中应用的研究成果做一整理,以期对后续的研究提供帮助。

关键词: 催化裂化, 集总动力学, 神经网络, 人工智能

Abstract: Fluidic catalytic cracking (FCC) is a complex process affected by many highly non-linear and interrelated factors, including properties of raw oil and regenerated catalyst, as well as the operating conditions of reaction. Mathematical modeling and analysis of the process and product yield optimization has been a hot research field in the petroleum processing. Lumped dynamic model is the most commonly used method in mechanism analysis for FCC. The complex composition of raw materials and products can be classified into finite components, which can be used to further analyze the product distribution and its influencing mechanism. It is an important part in the process of building lumped dynamics model to select suitable and fast methods of parameter estimation and calculation. Intelligent algorithms, such as genetic algorithm, particle swarm optimization and simulated annealing algorithm, can overcome the problems that classical algorithms depend on initial values and are difficult to find global optimum to a certain extent. Moreover, they can ensure the convergence of the algorithm. Thus, intelligent algorithms play a great role in promoting the development of lumped dynamics model. In addition, some factors ignored by conventional lumped analysis method can be analyzed by constructing an artificial neural network (ANN) model among the properties of raw oil, regenerated catalyst, operation conditions and product distribution from the statistical point of view, the product distribution can also be further predicted by the ANN, which is a new and effective way to construct the catalytic cracking analysis model. In this paper, the existing research results on data intelligent algorithm and the applications of ANN in the construction of catalytic cracking process model have been reviewed in order to provide possible help for the future research.

Key words: fluidic catalytic cracking, lumped kinetic model, neural network, artificial intelligence

中图分类号: