D-FNN based soft-sensor modeling and migration reconfiguration of polymerizing process

A soft-sensor modeling method based on dynamic fuzzy neural network (D-FNN) is proposed for forecasting the key technology indicator convention velocity of vinyl chloride monomer (VCM) in the polyvinylchloride (PVC) polymerizing process. Based on the problem complexity and precision demand, D-FNN model can be constructed combining the system prior knowledge. Firstly, kernel principal component analysis (KPCA) method is adopted to select the auxiliary variables of soft-sensing model in order to reduce the model dimensionality. Then a hybrid structure and parameters learning algorithm of D-FNN is proposed to achieve the favorable approximation performance, which includes the rule extraction principles, the classification learning strategy, the precedent parameters arrangements, the rule trimming technology based on error descendent ratio and the consequent parameters decision based on extended Kalman filter (EKF). The proposed soft-sensor model can automatically determine if the fuzzy rules are generated/eliminated or not so as to realize the nonlinear mapping between input and output variables of the discussed soft-sensor model. Model migration method is adopted to realize the on-line adaptive revision and reconfiguration of soft-sensor model. In the end, simulation results show that the proposed model can significantly enhance the predictive accuracy and robustness of the technical-and-economic indexes and satisfy the real-time control requirements of PVC polymerizing production process. (C) 2013 Elsevier B.V. All rights reserved.