面向苯乙烯聚合反应过程机理建模的鲁棒参数估计方法

The styrene polymerization process is a typical complex chemical reaction process. There are many model parameters in the procedure of mechanism modeling, which are difficult to be predetermined. High-precision mechanism model is of great importance for the application of product quality monitoring, control and optimization. In order to derive the high-precision mechanism model which is consistent with the actual process, parameter estimation (PE) is used as the key technology to tune the process mechanism model by using industrial measured data. In this proposal, based on the basic problems of PE for the mechanism model of styrene polymerization process, the parameter selection, the design of robust high-precision PE problem and the convergence of solving method will be focused. With detailed structure of the process mechanism model, the analysis of parameter selection is firstly investigated by considering the sensitivity, the importance ranking and the estimability of the model parameters. And then, the design strategies of the robust high-precision PE problem, based on information entropy theory and similarity analysis of operating conditions, are developed to decrease the influence of uncertainties from measurement error distribution and model parameters. Considering the coupling relationships between equations and variables, the solving strategies for the large-scale PE problem are proposed by using clustering based decomposition and parallel calculation. Finally, the practical industrial process (styrene polymerization process) is used to demonstrate the effectiveness of the proposed methodology and the validation of the tuned model. It’s expected that mechanism model based PE theory and methods could be systemized for styrene polymerization process, and the mechanism modeling for styrene polymerization process could be effectively achieved on a firm base of model analysis.

苯乙烯聚合反应过程是一种典型复杂化工反应过程，机理建模时参数众多且难以预估。利用工业测量数据对该过程机理模型进行参数估计(PE)，得到与实际过程相一致的高精度模型，是实现产品质量监控、控制与最优化策略的技术关键。本项目面向苯乙烯聚合反应过程机理建模，针对其PE的参数选择性、精度以及求解收敛性等几个关键基础问题进行研究。首先，利用机理模型的结构化特征，综合考虑参数的灵敏度、重要性排序及可估计性，研究参数的选择性分析方法；考虑测量误差分布的不确定性与模型参数变化的不确定性，基于信息熵理论与工况相似度分析研究鲁棒PE设计方法；从方程与变量的耦合关系考虑，研究大规模PE的拆解与并行运算方法；最后，通过实际工业过程(苯乙烯聚合反应过程)进行方法有效性验证与模型有效性的在线测试。本项目研究适用于苯乙烯聚合反应过程机理模型PE的理论和方法，为苯乙烯聚合反应过程建模提供模型分析基础。

苯乙烯聚合反应过程是一种典型复杂化工反应过程，机理建模时参数众多且难以预估。利用工业测量数据对该过程机理模型进行参数估计(PE)，得到与实际过程相一致的高精度模型，是实现产品质量监控、控制与最优化策略的技术关键。本项目面向苯乙烯聚合反应过程机理建模，针对其PE的参数选择性、精度以及求解收敛性等几个关键基础问题进行研究。首先，利用机理模型的结构化特征，综合考虑参数的灵敏度、重要性排序及可估计性，研究参数的选择性分析方法；考虑测量误差分布的不确定性与模型参数变化的不确定性，基于相关熵与工况相似度分析研究鲁棒PE设计方法，显著地提高了参数估计的精度；针对苯乙烯聚合反应过程的牌号变换时工况发生变化，构造了一种及时学习求解方法用来求解不同工况下的参数估计问题，该方法拟根据贝叶斯推理各工况之间的相似度概率大小，然后采用及时学习的方法构造当前运行工况的参数估计问题最好的初值点，有效地减小工况变化的不确定性对参数估计结果准确性的影响；针对大规模多模型系统对应的故障与正常操作条件下模型参数不同的特点，提出了一种基于粒子滤波多模型交互的故障检测与识别方法，该方法主要采用粒子滤波方法进行参数估计命题的动态优化求解，动态迭代求解过程采用多模型交互信息，从而得到准确的估计结果。考虑测量误差分布非高斯情况下，提出了一种鲁棒动态数据校正滤波方法，可提升扩展卡尔曼滤波器的状态估计的使用范围；从方程与变量的耦合关系考虑，研究大规模PE的拆解与并行运算方法，显著地提高了大规模参数估计问题求解的收敛率与快速性；最后，通过实际工业过程(苯乙烯聚合反应过程)进行方法有效性验证与模型有效性的在线测试。本项目研究适用于苯乙烯聚合反应过程机理模型PE的理论和方法，为苯乙烯聚合反应过程建模提供模型分析基础，并为其他工业过程的机理模型参数估计提供借鉴。