Abstract:

When modeling the oxidation reaction of p-xylene (PX), there are some differences between the reaction conditions of laboratory reaction equipment and industrial production processes. It is difficult to accurately describe the production status of industrial PX oxidation reactors through kinetic reaction models obtained in the laboratory. In the oxidation reaction process, the influence of various reaction operating conditions on the reaction process is mainly described by reaction rate constants. However, the relationship between reaction rate constants and various reaction factors is often uncertain, and even highly nonlinear. Therefore, machine learning methods (e.g. neural networks and support vector machines) are needed to obtain them. Due to the limited sample data provided by the laboratory, this paper proposed a weighted least squares support vector machine algorithm (BWO-WLS-SVM) based on beluga optimization for machine learning methods in small sample situations. Then intelligent optimization and correction were carried out on the parameters of the laboratory dynamics model, and an intelligent hybrid model that can accurately describe the PX oxidation reaction of industrial reactors was established, providing support for the optimization and control of the process.

Key words: p-xylene oxidation, weighted least squares support vector machine algorithm, beluga optimization, intelligent hybrid modeling

摘要：

对二甲苯（p-xylene，PX）氧化反应建模时，实验室反应装置及反应条件与工业生产过程有很大差异，这些差异导致了工业PX氧化反应器的生产状况很难通过实验室获得的动力学反应模型进行描述。在氧化反应过程中，主要通过反应速率常数来描述各反应操作条件对反应过程的影响，反应速率常数和各种反应条件之间经常存在非确定和非线性的函数关系，机器学习方法如神经网络或支持向量机等是解决该类问题的一种有效手段。此外，因为实验室提供的数据样本很少，针对小样本情况下的机器学习问题，本文在实验室机理和数据基础上，提出了基于白鲸优化的加权最小二乘支持向量机算法（BWO-WLS-SVM），并对实验室动力学模型参数进行了智能优化修正，建立了一个能够较为精确描述工业反应器的PX氧化反应智能混合模型，为该过程的优化及控制等提供了基础。

关键词: 对二甲苯氧化, 加权最小二乘支持向量机, 白鲸优化算法, 智能混合建模