Research And Application On Process Monitoring Based On Manifold Learning Method

With the development of modern technology,the procedure of industrial process is becoming more and more complex,the requirement of continuous,high efficiency and safety for process control becomes increasingly higher.The abnormal situation in the production being found timely and effective becomes more and more important.The method of process monitoring is an effective way to solve the above-mentioned problem.Among it,based on data driven has been widely applied to the field on fault detection and diagnosis.In this paper,on the basis of previous work,we carried out the research and application based on modified kernel supervised locally linear embedding(MKSLLE)and multi-manifold kernel semi-supervised(KSS(LPP-PCA))algorithm.The main research contents are as follows:(1)To solve some defects and deficiencies existing in the kernel supervised locally linear embedding algorithm,this article is improved on the basis of the above algorithm.On the one hand,it can solve the problem of so-called "out of sample" effectively and improve the generalization ability of data;at the same time,for the problem about the difficulty of selecting neighbourhood.On the basis of the traditional method,in this paper,we utilize supervised learning and local kernel principal component analysis to ameliorate the neighbor parameter which can more meet the practical requirement.(2)Considering the difficult problem about obtaining tag samples existing in the supervised methods,in order to fully develop potential information of unmarked data,in this paper,we propose process monitoring method based on multi-manifold kernel semi-supervised which can effectively utilize the unknown label sample to guide the learning process.In addition,this method is not only preserving the local structure information of data,furthermore,it fully considers the global diversity of sample information,depicted the global relevance of sample commendably.At last,we construct the optimal objective function for process monitoring.Finally,we use the proposed algorithms in this paper for the process of electro-fused magnesium furnace to simulate and analysis.Test their performance of fault detection and diagnosis respectively.Compared with some traditional methods,we verify the feasibility and effectiveness of the proposed method in this thesis.