Research On Industrial Process Fault Monitoring Based On Unbalanced Dat

With the rapid development of information technology,a large amount of process data can be collected and stored.Therefore,data-driven methods have become popular in dustrial applications.However,in real industrial processes,large amount of normal data should be regarded as the majority,and a few number of fault data should be regarded as the minority.Most classification methods can not perform well on imbalanced data.At the same time,the quantity of various fault data may be quite different from each other,which will further increase the complexity of the fault classification problem.Based on the existing research,this paper proposes improvements to the three classification methods of Bayes,Support Vector Machine(SVM)and Self-organizing Feature map(SOM).The D-S(Dempster-Shafer)evidence theory is used to realize the complementary advantages of multiple methods,and the fault simulation is realized based on TE imbalancef data.The main research contents of the paper are as follows:(1)Research on Bayes based on imbalanced fault monitoring.The basic theory of Bayes is elaborated in detail.Aiming at the problem of Bayes’ poor classification performance under imbalanced data,from the perspective of changing the sample distribution,the training set is processed by oversampling SMOTE(Synthetic Minority Oversampling Technique)and undersampling Easy Ensemble.At the same time,considering the limitations of these two resampling methods,a data regrouping technique T threshold Kmeans is used to process the training set.Through the comparison of simulation results,the advantages and effectiveness of the Kmeans-Bayes in imbalanced fault monitoring are verified.(2)Research on SVM based on imbalanced fault monitoring.The basic theory of SVM is elaborated in detail.Aiming at the problem of poor classification performance of SVM under imbalanced data,from the perspective of feature extraction,Principal Component Analysis(PCA)and Kernel Principal Component Analysis(KPCA)are used to extract features from data samples.At the same time,considering that the SVM sub-classifier still has a poor recognition rate for certain categories,an adaptive boost classifier(Ada Boost)is constructed to replace it.Through the comparison of simulation results,the advantages and effectiveness of the Ada Boost-SVM in imbalanced fault monitoring are verified.(3)Research on SOM baesd on imbalanced fault monitoring.The basic theory of SOM is elaborated in detail,KPCA is used for feature extraction of data samples and combined with mixed sampling,and SOM is used to establish a fault monitoring model.The clustering and visualization of data samples are realized through the SOM map,and the advantages and effectiveness of the KPCA-SOM method in imbalanced fault monitoring are verified through the simulation results.(4)Research on information fusion based on imbalanced fault monitoring.The basic concepts and fusion rules of D-S evidence theory are elaborated in detail.Kmeans-Bayes,Ada Boost-SVM and KPCA-SOM are selected as three fault monitoring models,and information fusion is realized according to Dempster fusion rules to obtain the final classification results.The simulation results verify the advantages and effectiveness of the D-S fusion in imbalanced fault monitoring.