Soft Sensor Of Polyvinyl Chloride Particle Properties Based On Fuzzy Neural Network

Particle properties of polyvinyl chloride are the important factor in processing and using performance. Because vinyl chloride polymerization process has the characteristics of strong nonlinearity and coupling, prediction for particle properties of polyvinyl chloride using mechanism modeling is difficult. In this article, soft sensor for particle properties of polyvinyl chloride using data-driven technique is developed. Aimed to enchaning model performance, a modeling method based on rough set-dynamic fuzzy neural network is proposed, and applied to predict the average particle diameter of polyvinyl chloride.This article mainly includes the following:(1) Firstly, the background of intelligent modeling technology and the basic element of intelligent modeling technology are introduced, including principal analysis, partial least squares, neural network, support vector machine and rough set, etc. Then these intelligent modeling techniques and its application in the modeling, optimization and controlling in industrial fields are described.(2) Three common fuzzy neurons of fuzzy neural network are analyzed. A dynamic fuzzy neural network is proposed to avoid the limitment of common fuzzy neural network. The characteristics, structure and algorithm of dynamic fuzzy neural network is detailly analyzed.(3) When dynamic fuzzy neural network is analyzed, the determinment of fuzzy rules need a long times and leads to the enchancment of the uncertainty of predicted results. A modeling method based on rough set-dynamic fuzzy neural network is proposed. Firstly simple of the decision table is used by the knowledge of simple in rough set theory. The simplest secondary variables are getted by all secondary variables are analyzed using the rough set membership function. In addition, a training algorithm based on simulated annealing method is propsed avoiding the lackment of BP algorithm.(4) Soft sensor for the average particle diameter of polyvinyl chloride based on rough set-dynamic fuzzy neural network is established. Firstly, the secondary variable of the average particle diameter of polyvinyl chloride is determined using the rough set membership function. The sample data is discrete by trisection method. The fuzzy rules are determined using the simpility of knowledge. At last, soft sensor of average particle diameter of polyvinyl chloride based on rough set-dynamic fuzzy neural networkis builted. The result indicated that the model based on rough set-dynamic fuzzy neural network model has better prediction accuracy, as the conventional dynamic fuzzy neural network.A modeling method is built by combining the rough set theory and neural network and applied to predict particle properties of polyvinyl chloride. The research result is helpful for controlling quality of polyvinyl chloride and improvement of the production and operation level.