The Semi-supervised Fuzzy Clustering Method Of Modeling The Quality Grade Of Cement Clinker

With the new dry cement production technology of modern cement plant growing popularity, it is imperative to effectively control the quality of cement products, in order to effectively save costs and improve the quality of clinker. At the present stage, the most optimization strategy of cement clinker still stay in the human experience, this paper apply the clustering analysis to the quality of cement clinker, attempt to find the intrinsic relationship between the quality grade of cement clinker and the parameters of production process, so as to lay the foundation for the optimal control of clinker quality.At the present stage, there are many ways to predict the product quality, based on thousands of real-time data which are collected from the DCS monitoring of modern factory, the neural network modeling,the fuzzy rules modeling and the fuzzy clustering modeling are selected as the preferred modeling method. But if the speed of modeling and the empirical knowledge are required, the semi-supervised fuzzy clustering model has a distinct advantage, it can be able to distinguish between monitoring parameters which closely related to the clinker quality and also take into account the experience knowledge to dynamic real-time correction the algorithm.This paper makes a detailed study of the clustering method of modeling the quality grade of clinker, then unsupervised FCM, semi-supervised dimensionality reduction and semi-supervised are used to distinguish between quality grade of clinker, and finally turn the quality grade of clinker algorithm into the form of rules. The key point of modeling techniques is that:the dimensionality reduction algorithm and the clustering algorithm. The key algorithms are:the fuzzy clustering algorithm and the semi-supervised FCM algorithm. The key control parameters are:the data optimization, the data filtering technology, the label selection method and the proportion parameter of monitoring information. On the specific algorithm, two monitoring information types are introduced to improve the unsupervised fuzzy clustering:One type is the pairwise constraints, the purpose is to reduce dimension of the data and improve similarity of the spatial. The other type is the labeled samples, the purpose is to initialize the cluster center and make some modification of the objective function. Experimental on real production data demonstrate that the improved algorithm can effectively improve the predict accuracy, reduce the clustering dimension and the computation time.The main contents are as follow:(1) The development actuality of data mining, semi-supervised clustering algorithm and quality of cement clinker is analyzed.(2)The fuzzy clustering theory is researched and a generalized fuzzy clustering model is given.(3)On the basis of the new dry cement production technology, statistical analysis is used to obtain the standard of quality of cement. The correlation analysis and the neural network analysis are used to extract a series of characteristic variables related to the quality grade of cement, on that basis, data mining tools are used to complete the work of the original data filtering.(4)The unsupervised fuzzy clustering algorithm is researched. In unsupervised FCM algorithm process, optimization function is used to complete iterative optimization of parameter values c and m, and the known sample points are joined to determine the value of clustering results. Finally the improved cluster algorithm is used to analysis the quality of cement. Results show that:By the unsupervised clustering alone, it cannot make a good distinction between the category 2 and category 3, it also cannot find all the partial information between the individual categories.(5)In order to overcome the shortcomings of the above clustering algorithm, this paper will introduce two types of monitoring information:Pairs of constraint are designed to construct the projection matrix, so as to change distance function and improve the iteration efficiency. Labels data are designed to change initial parameters and objective function, so as to guide the clustering process. In the end, in order to analyze the structure of model, the final model of fuzzy rules is built on the basis of FCM algorithm.(6)According to the actual production of cement, this paper analyze the impact of reduce dimension and proportion parameter of monitoring information. On the base of selecting optimal control parameters, the MATLAB program of semi-supervised fuzzy clustering model is designed. Experimental results show that:Compared to unsupervised FCM, semi-supervised FCM algorithm has been improved the output curve in the model, the forecast accuracy, the iteration time and the cluster spatial structure. It can find the partial information between the individual categories, can very well distinguish the difference between the categories, and also improve the details of the data distribution.