Optimization Of MI Soft Sensof For Propylene Polymerization Process Based On Fuzzy Neural Network

Polypropylene (PP), as one of the five most important plastics in the modern world, plays a significant role in our everyday lives. The PP product industry is the important component in the chemical industry, while the soft sensor research of the Melt Index (MI) is critical to the quality control of PP product. This paper is mainly focus on the soft sensor modeling of the MI by using the fuzzy neural network (FNN) method. Then the Support Vector Machine (SVM) and a novel self-organization mechanism are introduced to optimize the parameters in the FNN structure, respectively. The full analyses on the performance on the practical data from real industrial process prove that the optimized soft sensor model can well deal with the model structure problems and improve the performance of the soft sensor model. The several models brought in this paper can be applied in the practical industrial plant and serve as the options when dealing with the MI soft sensor modeling problems.The main works in this article are listed as below:(1) Get the directly detected variables from the PP industry and build up the MI soft sensor model by using the fuzzy neural networks, which combine the fuzzy system and the artificial neural network. The model firstly fuzzifies the input variables and then introduces the fuzzy rule and the defuzzification method to get the final output. The research result shows that FNN soft sensor is more effective than the traditional artificial neural network.(2) Optimize the basic FNN soft sensor based on the SVM method and change the expression to get the input that SVM can handle. Because SVM method can deal with the balance between the precision and the generalization ability, comparing with the traditional machine learning methods. This helps the soft sensor model to process the industrial noises in the sampled input datasets.(3) As to the difficulty related to the structure optimization problems when processing the industrial dataset, this paper gives a self-organization mechanism to adaptively change the structure of the soft sensor model. The mechanism uses2thresholds to decide the addition and deletion of the fuzzy rules in the FNN model, while the mechanism adds new fuzzy rules and deletes the useless fuzzy rules to better optimize the FNN soft sensor models.