MI Soft Sensor Method Research For Polypropylene Based On SFA And DBN

Melt index as one of the most important quality variables in polypropylene plant,is difficult to measure online and can only be obtained by artificial sampling and laboratory analysis.Studies on soft sensor technology of melt index have important theoretical and practical value on optimization of polymerization process and control of product quality.This paper focuses on soft sensor modeling of polypropylene melt index.This paper studies on the soft sensor model of nonlinear slow feature regression,which improves the prediction precision of the model.The extreme learning machine is applied to the training process of deep belief network to establish the soft sensor model based on DBN-ELM,which improve the performance of the deep belief network.This paper also studies on the dynamic soft sensor modeling based on IDBN-GM.The GM(1,1)is applied to the dynamic calibration of static model to establish the dynamic soft sensor model.The main research works are summarized as follows:The soft sensor model based on nonlinear slow feature regression is presented.Firstly,the slow feature analysis is extended to the nonlinear form,which is used to extract the slow features.Finally,the regression model is established based on the extracted features,which is used for the prediction of polypropylene melt index.The validity of the model is verified through industrial data sets.The model based on DBN-ELM is studied for the low prediction precision and multi-grade polypropylene process.Deep belief network as a deep neural network model,it has excellent feature extraction ability,which makes it easier for classification or prediction.In addition,extreme learning machine as a kind of fast learning algorithm for neural network with single hidden layer,it is applied to the training of the last layer of deep belief network to improve the performance of the model.The simulation results of industrial data indicate that the presented model has better prediction accuracy than traditional soft sensor models.A new kind of dynamic soft sensor model based on IDBN-GM is presented for the dynamic characteristics of industrial process.Firstly,the static model is established based on DBN-ELM,of which the results are compared with the actual values.Thus the errors sequence is obtained and modeled and predicted by the GM.Finally,the predictive error results are combined with the static model to realize dynamic correction.The simulation results of industrial data indicate that the presented model has better prediction accuracy than single DBN and DBN-ELM model.