Research On Soft Sensing Of Free Calcium Oxide In Cement Clinker Based On Resnet And Long Short Memory Network

The content of free calcium(fCaO)in cement clinker is an important factor affecting the quality of cement,which is directly related to the stability of cement and the energy consumption of production.At present,the content of fCaO in cement clinker is measured off-line mainly through laboratory analysis in China,but this method has obvious lag in the control of cement firing system.In order to achieve the goal of stable production,energy saving and consumption reduction,this topic combines the actual situation of cement production,takes the content of cement clinker fCaO as the research object,based on the time series of deep residual network(TS-Resnet)and principal component analysis method The memory network combined(PCA-LSTM)method is used to study the soft-sensing scheme of cement clinker fCaO content,so as to realize the model establishment and real-time prediction of clinker fCaO.The main research contents of this article are as follows:1.We study on cement firing process and clinker fCaO generation mechanism,analyze the main factors affecting the fCaO content of cement clinker,select variables related to clinker fCaO content as auxiliary variables of the soft measurement model;use filtering,normalization and other methods The data of each auxiliary variable was preprocessed.2.Aiming at the characteristics of time-varying delay,non-linearity and uncertainty among the variables during the cement firing process,the time series of multiple auxiliary variables affecting the cement clinker fCaO are selected as the model input,and the depth residuals based on the time series are established The internet.Through the model's learning of the cement data characteristics,the convolution and pooling of the residual module are used to perform deep multiple feature extraction on each process variable.The quick connection in the residual module effectively avoids the gradient dispersion problem during training and improves the prediction accuracy and generalization of the model are introduced.3.Due to the problems of high auxiliary variable dimensions,large data volume,and different sampling frequencies during the cement firing process,a soft measurement method based on the combination of principal component analysis and long-term and short-term memory networks was proposed.First,the high-dimensional data is reduced to the target dimension using PCA,and then the dimensionality-reduced data and the remaining low-dimensional data are fused to the LSTM for training,and the data features are extracted to predict the clinker fCaO content.4.Based on the above model and related theoretical analysis,using C # programming language,Python programming language and SQL Server 2014 database,using VS2017 as a platform,designed and implemented a real-time prediction software for cement clinker free calcium.Afterwards,the on-site data were used to experiment and analyze the model proposed in this paper,and compared with various existing mainstream soft-sensing models in terms of training time,training prediction effect and other indicators.The results show that the real-time prediction scheme of clinker fCaO mentioned in this paper is general strong ability and high accuracy.