Modelling and Simulation of Polypropylene Melt Index Based on Stacked Neural Networks

Modelling and simulation of polypropylene melt index based on stacked neural networks is presented. Enhancing model accuracy and robustness using multiple neural networks is studied. Instead of developing a single neural network estimator, several neural network estimators are developed and combined together to improve estimation accuracy and robustness. Combining individual neural networks using ridge regression is proposed. For the purpose of comparison, single neural network models, and stacked neural network models are developed and evaluated. The estimation errors can be further reduced by using stacked neural networks. The application of the proposed modelling method to the development of soft sensors in an industrial polypropylene polymerization plant demonstrates its effectiveness.