The Research Of Soft Measurement Of The Free Calcium Oxide In Cement Clinker Based On Time-Series Attention

The free Calcium Oxide(f-Ca O)content of cement clinker is a key indicator of cement quality and is one of the most important parameters for cement production control.Currently,the detection of clinker f-Ca O content mainly relies on manual chemical off-line sampling and analysis,however,off-line testing has a large time lag and is not conducive to real-time monitoring of cement quality.This paper takes the soft measurement method of clinker fCa O content as the research object and proposes a soft measurement model of clinker fCa O content based on time-series attention for the real-time measurement of clinker f-Ca O,the specific research is as follows:Firstly,the production process of new dry process cement and the process mechanism of clinker firing are studied,the role of process equipment and the key factors affecting the f-Ca O content of clinker are analyzed in depth,and corresponding solutions are given by studying the inherent characteristics and modeling difficulties of the variables in the clinker calcination process.Secondly,in order to effectively select process variables and reduce data redundancy,a variable selection method based on the fusion of maximum average mutual information theory and cement process mechanism analysis was used.The method firstly selects the relevant variables affecting the f-Ca O content initially through the analysis of the firing process of clinker,and then calculates the correlation coefficients between the variables using the maximum mean mutual information method,and selects the process variables with high contribution to the f-Ca O of clinker in the second stage.Then,a network model(DE-CAM-CNN)based on data enhancement and continuous attention mechanism is developed for the nonlinear,strongly coupled and time-varying time delay characteristics of the clinker calcination system.The data enhancement and continuous attention methods were used to focus on the continuous action time duration information of the process variables;then the CNN(Convolutional Neural Networks)network model was used to deal with the multivariate temporal coupling relationship and extract the key features affecting the f-Ca O content of clinker.Experiments show the effectiveness of the method and the high measurement accuracy of the model.Finally,a network model(ADM-WGM-CNN)based on data decoupling method and window selection mechanism is established to address the characteristics of multiple working conditions and dynamic changes of clinker firing process.The data decoupling method is used to eliminate the coupling relationship between variables,and the adaptive window gating mechanism is used to dynamically select the duration of variables.Experiments show that the method is effective,and the adaptive algorithm further improves the measurement performance of the network model.