部分状态变量不可测的乙腈精制过程智能建模

Intelligent Modeling of Acetonitrile Refining Process with Partially Unmeasurable States

提出一种基于PLS与NN的混合智能建模方法，建立部分状态变量不可测的高纯乙腈连续精制过程模型。针对乙腈连续精制过程工况变化频繁、部分关键状态变量不能在线检测且难以实现基于机理模型的过程综合优化控制等难点，并考虑到过程的强非线性、多变量关联性及小样本问题，利用PLS进行相关性分析与主成分提取，解决变量之间的相关性及小样本多变量问题，并利用优化程序完善模型训练过程，提高建模精度。预测结果表明，PLS-NN混合智能模型与NN模型相比，具有更好的预测性能与泛化能力，证明了方法的有效性。

A hybrid intelligent modeling approach based on partial least-squares （PLS） regression and neural networks （NN） was developed to predict the partially unmeasurable states of continuous refining process of high purity acetonitrile. According to the difficulties of varying operation condition, dynamic process with partially immeasurable states, unablilty to realize the integrated optimal control based on mechanism model, and considering the problems of small sample, non-linear and multi-variable coupling, this modeling approach extracted relatively few latent variables as inputs of intelligent model by using correlation analysis and principle components extraction. Then a constructive approach and program was used to optimize training process and improve the precision of modeling. The simulation indicated that PLS-NN model outperformed the normal neural networks on prediction and generalization, which proved the validity of this approach.