聚合反应过程质量指标的推理估计混合模型

A Hybrid Model for Inference Estimation of the Performance Figure in Polymerization Process

针对聚合反应过程的非线性、时变性和不确定性,提出了一种多类型混联混合推理估计模型。该模型以过程机理知识为基础框架,以各种神经网络和回归辩识模型的计算结果作为混合模型中各子模型或机理模型的过程参数。为了体现过程的多模式集成特点,该混合模型充分利用各种类型模型的不同特性,既保证按照动力学规律描述聚合反应过程特性,又充分利用现场运行和分析的数据,辩识模型结构参数,使所建模型不必完全依赖对过程特性的认识。将该混合模型用于聚丙烯腈生产过程质量指标的推理估计,现场应用效果证明了这种模型的优良性能。

In order to overcome the difficulties induced by the nonlinearity, time-variation and indeterminacy of polymerization process, a manifold series-parallel hybrid model for inference estimation was presented, which was based on process mechanism knowledge and the results obtained from neural networks models and regression models. The process parameters of the hybrid model can be extracted from those results mentioned above. In order to express multi-mode integration in polymerization process, the hybrid model was derived by means of the dynamic characteristics of different models. As a result, not only the correspondence between the characteristics description of polymerization processes and kinetics can be guaranteed, but also the structure parameters can be identified by the practical operation and analytical data. This lead to the fact that the model not entirely relied upon the mechanism knowledge of the process. By using the manifold series-parallel hybrid model, different dynamic characteristics in polymerization processes can be revealed. The model has been used in polyacrylonitrile production for performance figure estimate, and it has been verified that the effectiveness is extremely obvious.