Research On Soft Sensing Method Of Cement Raw Material Fineness Based On Mutual Information And Support Vector Machine

Raw material fineness is an important indicator to assess the quality of raw material grinding.It is one of the keys to the control of the raw material grinding system and the entire cement production system to ensure the quality and stability of raw meal fineness.However,at present,the method of testing is mainly used to detect the fineness of the raw material off-line.This method has obvious time lag,there is a tendency for excessive grinding of the raw material,and the quality and economic benefits cannot be taken into account.This paper proposes a soft measurement method based on the mutual information and support vector machine to realize real-time online forecasting of raw material fineness.The specific research work is as follows:Firstly,the operation mechanism of the cement raw material grinding process is studied,and the influencing factors affecting the quality of the raw material grinding and stable operation were analyzed.Combining the research status of raw material soft grinding measurement at home and abroad,a modeling strategy of raw material fineness soft measurement is proposed,and candidate input feature variables are initially obtained.It lays the foundation for the establishment of a soft measurement model for cement fineness.Secondly,according to the raw material grinding process,it has the characteristics of multi-variable coupling,strong nonlinearity,and time delay.The k-nearest neighbor mutual information method is used to characterize the degree of correlation between high-dimensional variables,and the time delay characteristics of the raw material grinding system are selected based on the maximum correlation criterion.A method combining sequential forward selection(SFS)and backward sequence selection(SBS)is proposed.This method is used to select the input features,and the input variables of raw material fineness soft measurement model are obtained.Thirdly,The presence of sample data skew and sample data redundancy increases the complexity of the algorithm and reduces the performance of the model.To solve this problem,this paper proposes a KS(Kensard-Stone)sample reduction method.And based on T~2 test and Bartlett test to evaluate the effectiveness of the reduction method.Least squares support vector machine(LSSVM)is used to train the reduced sample and establish a soft measurement model of raw material fineness.Finally,the idea of combining mutual information,KS and LSSVM was put forward,and the soft measurement model of raw material fineness based on MI-KSLSSVM was established.Field data were used for experiment and analysis.The results show that the proposed method has fast convergence and strong generalization ability.It can realize real-time prediction of raw material fineness.