Application Research On Intelligent Control Method Of PVC Manufacturing Process

PVC production process is a dynamic process with matter transmission, heat exchange and complicated chemical reaction. From the control view, PVC production process is a kind of typical and complex industrial process due to its nonlinearity, strong coupling, slow time-varying and large time delay. A great deal of uncertainty information and multiplex data make it difficult to control PVC production process effectively by using conventional control methods. The intelligent control theory provides an efficient approach to realize the control of this kind of complex industrial process.Based on the review of the related research references, the detailed investigation on the mechanism and technique and the comprehensive collection of history data, expert experience and manipulation regulations of PVC production process, an integrated modeling and intelligent control strategy of PVC production process is proposed by comprehensive utilization of PVC producing process, computer control technology, soft-sensor modeling technique, modern control theory and intelligent control technique. The porposed strategy mainly contains polymerizer rough set-neural network fault diagnosis algorithm based on improved discernibility matrix attribute reduction, polymerizer process soft sensor modeling method based on T-S fuzzy neural network optimized by harmany search algorithm, the generalized predictive algorithm of polymerizer temperature based on segmental affine and adaptive decoupling control method of PVC stripping process based on neural network.This dissertation has carried on the following research:(1) Aiming at the real-time fault diagnose and optimized monitoring requirements of the large-scale key equipment polymerization of PVC production process, a real-time polymerize fault diagnose strategy is proposed based on rough sets (RS) theory with the improved discernibility matrix and back propagation (BP) neural networks. The improved discernibility matrix is adopted to reduct the attributes of rough sets in order to decrease the input dimensionality of fault characteristics effectively. Then Levenberg-Marquardt BP neural network is trained according to the reducted decision table in order to decide the configuration parameters of the proposed polymerize fault diagnose model. Thus the classification of the fault patterns is to realize the nonlinear mapping from fault symptom set to polymerize fault set according to a set of symptoms. Polymerize fault diagnose simulation experiments are carried out combining with the industry history datum. Simulation results show the effectiveness of the proposed fault diagnoses method based on rough set and BP neural networks.(2) Directing at the question of hard real-time access and being difficult to achieve qualitative close-loop control of VCM’s conversion rate and velocity in PVC polymerizing process, a soft sensor modeling technology of VCM’s conversion rate and velocity is proposed based on multi T-S fuzzy neural network model combining on fuzzy c-means (FCM) clustering algorithm. Firstly, the principal component analysis (PCA) method is adopted to select the auxiliary variables of soft-sensing model in order to reduce the he model dimensionality. Then a hybrid optimization algorithm utilizing the harmony search (HS) and least square method are proposed to optimize the structure parameters of T-S fuzzy neural network. Simulation results show that the model can achieve real-time forecasting and monitoring on conversion rate and velocity, and have great significance to impove quality and yield of PVC product.(3) According to the characteristic of nonlinear and difficulty to control the temperature model, a generalized predictive control algorithm of PVC polymerization reactor temperature based on segmental affine is proposed. Firstly, the dynamic equation of polymerizer temperature is derived in accordance with heat balance mechanism. Segmental affine model is achieved on the basis of polymerizer temperature and conversion. State space of each subsystem is described by oval set. Next, LIM is used as solution tool to design the controller, and is used to structure Lyapunov function so as to analyse the stability of system. Finally, this segmental affine model is applied in the generalized predictive control algorithm. Simulation results show the validity and feasibility of the proposed algorithm.(4) In the light of stripper’s nonlinear, strong coupling and time-varying characteristics, the adaptive decoupling control algorithm of PVC stripping process based on neural network is proposed. Firstly, the controlled object model of the stripping process is established based on the data-driven dynamic fuzzy neural network (D-FNN) method. Then the decentralized neural network controller is adopted to decouple the stripping process into the two single-input-single-output (SISO) of slurry flux versus tower top temperature and steam flux versus tower bottom temperature. Finally, the BP neural network PID controller is used to control the decoupled SISO system. Simulation results show the validity of the proposed integrated control strategy.(5) The above intelegent modeling and optimized control methods are applied to PVC production process. The results show that the propose strategy can not only improve quality and yield of PVC product, and eliminate environment polution, but also promote to increase overall cost-effectiveness of PVC. Meanwhile, practical methods are provided for intelegent control in the other complex industrial process.