Abstract: Aiming at problems appearing in model switching of multi-model predictive control,a multi-model switching method based on recognition of operating conditions was presented,and the variations of dynamic characteristics in the process reflected by measurable variables were adopted.Firstly,Gaussian mixture model(GMM)was used to classify historical data into several operating conditions. Then the load vector matrix and the predictive models were built based on the data of different operating conditions. Lastly,the predictive model was chosen according to squared prediction error(SPE)of the principal component analysis(PCA)model.This method was implemented in controlling the coil outlet temperature(COT)in an ethylene pyrolysis furnace.The simulation results showed that the stable and balanced control of multiple COTs were realized by the presented method,and the matched operating condition could be easily found by comparing SPE of different PCA models when the operating condition of the system varied.Meanwhile,the corresponding predictive model was selected,with which the problem of switching lag among predictive models when the dynamic characteristics of the system changed was resolved.

Key words: multi-model switching method, operating condition, Gaussian mixture model, squared prediction error, multi-model predictive control

摘要： 针对多模型预测控制的模型切换问题,提出了一种基于工况判断的多模型切换方法,利用工业过程中的可测变量综合反映系统的动态特性,根据动态特性的变化进行多模型切换。首先利用高斯混合模型(GMM)将历史数据划分为若干个工况,然后利用不同工况下的历史数据建立负荷向量矩阵和预测模型,最后根据主元模型的平方预报误差(SPE)选择预测模型。以乙烯裂解炉的反应管出口温度(COT)的控制为例进行仿真,仿真结果表明:提出的方法实现了多个反应管出口温度的稳定均衡控制,当系统的工况发生改变时,通过不同主元模型的SPE统计量的比较,可以很容易地找到匹配的工况,并切换为相应的预测模型,解决了当系统动态特性发生改变时,预测模型切换滞后的问题。

关键词: 多模型切换, 工况判断, 高斯混合模型, 平方预报误差, 多模型预测控制