Abstract:

C₃ hydrogenation process has variable operating regions and cascade reactions. The conventional manual hydrogen operation is prone to over-hydrogenation or alkyne leakage, resulting in large fluctuations in the outlet methylacetylene (MA) and propadiene (PD) concentrations and also affecting the catalyst selectivity and conversion ratio. Due to the nonlinear features and changeable operating regions of the C₃ hydrogenation process, traditional model predictive controllers based on linear models have great limitations in industrial process control. The Hammerstein model is a nonlinear model containing a nonlinear steady-state part and a linear dynamic part, which is suited to describe nonlinear industrial processes. In this work, a robust minimum covariance constrained control method is proposed for C₃ hydrogenation process control based on the Hammerstein model. The nonlinear steady-state part of Hammerstein model is approximated by WaveARX neural network with multiple operating region data and the linear dynamic part is identified according to the first principle analysis with the operation data. In this work, the nonlinear robust control problem is transformed into a linear model robust control and nonlinear model inversion problem. Based on the Hammerstein modeling bias, the upper bound limit of output bias covariance is calculated and minimized to get the sub-optimal state feedback control law, and then obtain the control input of the system according to the inverse model of the nonlinear part. The simulation and industrial application results have verified the feasibility and effectiveness of the proposed method.

Key words: hydrogenation, process control, Hammerstein model, neural network, minimum covariance constrained control, linear matrix inequality

摘要：

碳三加氢反应器操作条件多变，并且存在级联反应，传统的手动操作配氢容易过加氢或漏炔，导致出口甲基乙炔(MA)和丙二烯(PD)浓度波动较大，还会影响催化剂的选择性和转化率。由于碳三加氢过程具有非线性、操作区域多变的特点，传统的基于线性模型的模型预测控制器在工业过程控制中存在较大的局限性。Hammerstein模型是一种将非线性稳态环节和线性动态环节串联的非线性模型，非常适合描述非线性工业对象且方便控制器设计。提出了一种基于Hammerstein模型的鲁棒最小协方差约束控制方法，Hammerstein模型的非线性稳态部分采用WaveARX神经网络逼近，并根据多操作点稳态输入输出数据进行辨识，通过机理分析辨识得到的操作点附近线性化模型作为线性动态部分，将非线性鲁棒控制问题转化为线性模型鲁棒约束控制和非线性环节求逆问题，基于Hammerstein模型的建模偏差，对模型输出偏差协方差上限进行约束并最小化，采用线性矩阵不等式求解出次优状态反馈控制律，再根据非线性部分的逆模型得到系统的控制输入，仿真及工业应用结果证实了所提方法的可行性和有效性。

关键词: 加氢, 过程控制, Hammerstein模型, 神经网络, 最小协方差约束控制, 线性矩阵不等式