Abstract: Grade transition processes are critical for saving energy in polymerization, however, the traditional models do not pay much attention to the transitional process due to insufficiency of quality data. Therefore, based on the previous work of the three-stage decomposition modeling method, the overall process is further dealt with. In the polypropylene process, there are stable production processes for a single grade and grade transitional processes between two different grades. As these two kinds of processes have different dynamic characteristics, it is reasonable to divide data into different sub-domains according to different dynamic processes and develop sub-models in every domain. In this paper, construct ensemble of models based on different dynamic processes and a discriminant model to detect the beginning and completion of the transition were proposed. Finally, the proposed model was applied to industrial plant data and its higher prediction performance was verified, compared with a single model.

Key words: model, dynamic modeling, polymerization, three-stage model, grade transitional process

摘要： 牌号切换过程是聚合反应过程节能降耗的关键环节，已有的模型往往因为质量样本较少而弱化甚至避开这一过程。因此，在之前专门针对牌号切换过程提出的三阶段分解方法的基础上，进一步面向整个生产过程，针对聚丙烯反应过程存在的同一牌号的稳定生产过程以及不同牌号间的切换过程具有不同动态过程的特性，按照不同的生产模式和生产牌号划分不同动态过程的样本，采用多模型的方法在各自的样本集上建立子模型，有针对性地把握相应的动态变化规律。为了实现多个子模型之间的切换，进一步基于反应条件和反应结果估计值构建了综合判断模型。最后，通过实际数据验证，三阶段多模型相对Kim多模型、单一模型来说，具有更好的预测结果。

关键词: 模型, 动态建模, 聚合, 三阶段模型, 牌号切换过程