Soft Measurement Method Of Phenolic Resin Purity Based On Neural Network

As a common chemical raw material,free phenol is one of the important indicators for testing its purity.With the increasing use of phenolic resins,the requirements for their purity are getting higher and higher.Due to the large number of variables affecting the synthesis process of phenolic resin,the quality of the final product is difficult to be guaranteed.So far,there is no fixed method to measure its purity.Therefore,it is of great practical significance to use computer technology to predict the purity of phenolic resin.Based on this problem.Phenolic resin produced in a factory as the research object of this paper.Neural network and soft measurement technology as the theoretical basis.And a soft measurement model of phenolic resin purity is designed.Firstly,phenolic resin production process was introduced and analyzed.Combined with the actual experience of the field process personnel,14 measurable variables affecting the purity of the finished product were determined.And sample data collection and pretreatment were performed.Secondly,the principal component analysis of the processed sample data is carried out.And the 14 variables related to the purity of the finished phenolic resin are converted into 7 principal component variables which can fully represent all the information of the original data.Realizing the reduction of the original data set.Then,a soft-measurement model based on BP neural network phenolic resin purity is established.The modeling method is divided into the whole element method and the principal element method.The MATLAB software was chosen to write the program.The designed phenolic resin neural network structure was trained and simulated.Through the analysis and comparison of the relative error of the prediction results,it is concluded that the performance of the prediction model established by the elementary method is relatively better.Finally,the BP neural network model is optimized by standard differential evolution algorithm.The optimized DE-BP model is simulated and analyzed.The obtained training and test results are compared with the principal component prediction model established by BP neural network.The comparison results show that the optimization effect is good.The correction method of the soft measurement model is proposed at the end of the paper.In the actual application process,the soft sensor model will affect the prediction performance of the model due to the existence of external or unmeasured interference factors.The network structure and parameters are corrected by short-term correction and long-term correction.The prediction effect of the soft-measurement model is effectively guaranteed in the actual application process.