Research On Soft Sensor Modeling Of Vinyl Chloride Rectification Process Based On Neural Network

With the development of the domestic economy,polyvinyl chloride(PVC)now get a large-scale use in different application areas of our country,and its requirement is also growing in every day.The mass fraction of vinyl chloride monomer(VCM)is the main substance in the PVC polymerization process The mass fraction of its monomers affects PVC and its future development in the industry to varying degrees.Nowadays,many countries conduct in-depth research on the VCM rectification process,the purpose of which is to reduce the other substances in VCM to improve its purity.In particular,some means commonly used in production vinyl chloride makes monomers contain many organic and inorganic impurities,and there are many complex variables in the process of vinyl chloride rectification,which is hard to control VCM concentration.Therefore,controlling the rectification process of vinyl chloride must be a key part of improving the performance of PVC products.In order to solve the problem that there are many variable factors in the rectification process of vinyl chloride and it is hard to obtain the mass fraction of VCM online in real time,In this study the mass fraction of VCM was predicted by establishing soft sensor model.The main factors affecting the mass fraction of VCM are selected through the analysis of the rectification process of VCM.Because the neural network structure is too large or small,it will affect the output results,especially the structure parameters are obtained randomly,so the intelligent optimization algorithms are used to improve this structure,and the data is preprocessed before modeling.This paper proposes two soft sensor models for predicting the mass fraction of VCM during the rectification of vinyl chloride based on neural networks.The specific work content is as follows:First,the soft-sensor model based on LLE-WOA-RBF neural network is established.Use the Locally Linear Embedding(LLE)to perform dimensionality reduction mapping on high-dimensional input data,reconstruct the effective factor of the input data,and maintain the structure between the initial input data.Using the better optimization ability and faster convergence speed of the Whale Optimization Algorithm(WOA),the parameters in the radial basis function neural network(RBF)neural network are optimized.Thereby improving the generalization ability of RBF neural network and realizing the nonlinear mapping between input and output variables.Secondly,in order to further improve the overall prediction accuracy of the soft-sensor model.establish the soft-sensor model based on t-SNE-TSWOA-RBFNN.The t-SNE maps the influencing factor data to a low-dimensional space,and retains the primeval characteristics of the data to reduce the complexity of the soft sensor model.Combining whale optimization algorithm(WOA)and Tabu Search Algorithm(TS)to help the algorithm enhance its global search capabilities.The TSWOA algorithm is used to optimize the parameters of the RBF neural network model to realize the prediction of the concentration and quality of vinyl chloride.Finally,the simulation comparison experiment was performed on the above two soft-sensing models.The simulation consequences demonstrated that the two models have good prediction capability and generalization ability,can meet the conditions of real-time measurement of the mass fractionn of VCM in the process of rectification.And it will have better application prospects in complex industrial control processes.