Abstract: Molecular weight distribution (MWD) of polymer is an important performance index of quality.Due to the MWD can't be measured online,direct quality control is difficult to achieve.The combined of Legendre orthogonal polynomial and neural network had been used to build the model of the MWD,then the MWD tracking control in three-dimensionals space were decomposed the tracking control of its distribution moment in two-dimensionals time domain. The mothod,which employed the moments of distribution to predict the MWD,was proposed. The optimal objective function was based on the sum of squared errors (SSE) of the moments under certain constraints,meanwhile the correction term of the lower order moments was also introduced,in which part of the closed-loop feedback control of MWD can be obtained.The proposed control method was tested on styrene polymerization reaction in CSTR,and perfect control performance shows the effectiveness of the work.

Key words: complex neural network, polymerization reaction, molecular weight distribution, moment, model predictive control

摘要： 聚合物分子量分布（molecular weight distribution，MWD）是聚合产物重要的质量指标，由于无法在线测量，使得直接质量控制至今难以实现。在利用Legendre正交多项式组合神经网络建立聚合反应分子量分布灰箱模型的基础上，把MWD这个三维空间控制问题解构为以其矩向量为特征的二维时间域的控制问题，提出了通过控制分布的矩值实现分子量分布的预测控制方法。目标函数以矩值误差平方和为基础，考虑控制变量的约束条件，同时引入可测低阶矩的修正项，使得分子量分布的部分闭环反馈控制得以实现。该方法以实验室规模的苯乙烯聚合过程为对象进行了仿真建模与控制研究，获得良好的控制效果，证明了方法的有效性。

关键词: 组合神经网络, 聚合反应, 分子量分布, 分布函数矩, 模型预测控制