Complex Industrial Processes Based On Ls-svm Fault Diagnosis Method

Complex industrial process plays an important role in improving the national economy. Due to its complexity and difficulty in modeling, more attention is paid to fault diagnosis of complex industrial process. In recent years, a new machine learning algorithm, support vector machine (SVM), is widely applied in many fields. Least squares support vector machine (LS-SVM) can act a better performance than SVM in the case of large samples. Based on LS-SVM, fault diagnosis methods are researched deeply in this thesis.Firstly, several kinds of fault diagnosis methods are summarized. The characteristics of LS-SVM for classification are introduced and the fault diagnostic scheme based on LS-SVM is presented. Relevant examples are given to show the availability of fault diagnosis based on LS-SVM.Secondly, in order to improve the effectiveness of diagnosis, a diagnostic method based on fuzzy C-means (FCM) clustering and LS-SVM is proposed. The samples are clustered using fuzzy C-means clustering algorithm during the data preprocessing, and fuzzy theory is used in classific decision fuction to enhance the ability of classification. Satisfied results are achieved by using this method.Finally, in order to ensure the real-time requirement, an integrated method based on principal component analysis (PCA) combined with LS-SVM is proposed. In this method, wavelet transform is used to eliminate the noise in preprocessing, and then PCA model is established to monitor the process.When a fault is detected, LS-SVM is used to classify the fault accurately. This method combines the advantages of PCA and LS-SVM. By using this method, online fault monitoring is implemented and fault types can be classified successfully. The simulation results of TE process show the effectiveness of this method.