Research On Soft Sensor Method Of Free Calcium Oxide In Cement Clinker Based On Data-driven

Cement clinker free calcium oxide(fCaO)is an important indicator for assessing the quality of clinker calcination,and its accurate prediction has an important guiding role in optimizing production.At present,the method of on-site sampling and offline testing is mainly used to obtain the fCaO content value,but the fCaO content value obtained by this method has an obvious time lag for the control system.This paper discusses the problems of time series data and fewer label samples,and proposes fusion of multi-scale spatio-temporal features and bidirectional long-short-term memory network(MS-Bi LSTM),multi-scale spatio-temporal features and deep non-negative sparse autoencoding network(MS-DNSAE).A soft sensor model of clinker fCaO to realize real-time prediction of clinker fCaO content value.The specific research work is as follows:First,analyze the current research status of cement clinker fCaO soft measurement at home and abroad,and propose a clinker fCaO soft measurement modeling strategy;analyze the cement clinker fCaO calcination process and its generation mechanism,which the factors affects the stable operation of the cement clinker process and the content of clinker fCaO are discussed,the input variables of the soft sensor model are selected,and the 3? criteria,mean filtering,and data standardization are used to preprocess the selected variables.Secondly,the multi-variable time series data is analyzed in the time dimension using wavelet transform,and the signal is decomposed into signals of different time and space scales;for the characteristics of multi-variable coupling and strong nonlinearity in the clinker calcination process,Bi-directional long and short-term memory network(BiLSTM)extracts the coupling relationship between variables,selects nonlinear features based on the nonlinear activation function;proposes a soft sensor model based on MS-BiLSTM.Then,the sampling interval of clinker fCaO is one hour,and most of the process variables such as temperature,current,and voltage are collected in real time,and these process variables imply the law of cement production and the actual production status.Aiming at the problem of low accuracy of the soft sensor model caused by limited label samples,a soft sensor model based on MS-DNSAE was established.Firstly,non-negative sparse constraint autoencoder is used to obtain the hidden layer expression of the unlabeled data set,and then the label sample set is used for reverse fine-tuning to realize the prediction of clinker fCaO.Finally,using field data to experiment and analyze the proposed soft sensor model,the results show that the proposed soft sensor model has high accuracy and strong generalization ability,and can realize real-time prediction of clinker fCaO.