Research On Fault Diagnosis Method Of Steam Turbine Generator Based On Information Fusion Technology

Turbo-generator set as the core equipment of power system, whose normal running directly affects the safety and stability of the entire power system. The study on fault monitoring and diagnosis online of steam turbine generator, can effectively reduce the economic loss and risk caused by excessive maintenance. The traditional generator fault diagnosis system is based on a single or a single signal characteristics,due to the complexity of the motor structure and its runtime environment, usually some kind of fault may be accompanied by a variety of characteristics or some kind of fault characteristics may be caused by a variety of faults, thus the fault diagnosis technology based on a single characteristic has its limitations, which is difficult to meet the requirement of complex motor fault diagnosis. As the increasing amount of information monitored about generator, how to effectively utilize the various of comprehensive information to extract the fault feature of generator to reduce and eliminate the uncertainty of using a single fault information, to overcome the limitations of using single diagnosis method, to improve the accuracy of fault diagnosis is an important research direction and is also very necessary.In this thesis, the concept and the hierarchical model of information fusion technology and its application status in the field of fault diagnosis are firstly expounded; then, the failure mechanism of several common faults of generator is analyzed, the difference of the harmonic characteristics and vibration characteristics between stator and rotor when occurred different faults of a generator is obtained, and use which as the features of fault samples used for fault diagnosis; a multi-sensor feature fusion algorithm based on kernel principal component analysis and clustering of combining the mutative scale chaos characteristics is put forward, the KPCA algorithm is firstly utilized to extract the characteristics of the fault samples – the nonlinear kernel principal yuan, and selects the nonlinear kernel principal yuan as the fault characteristics of sample, then use the mutative scale chaos clustering algorithm for fault identification; an particle swarm optimization clustering algorithm is improved, which uses feature weighting method based on the sample similarity to get the weighted fault samples by weighted the different dimensions of characteristics, and use the improved particle swarm algorithm to obtain the optimal clustering center of all kinds of fault correspondingly, according to a recent distance principle to decide the fault type of each fault sample, which can achieve a variety of fault diagnosis based on vibration characteristics; finally, the improved D-S(Dempster-Shafer) evidence theory based on the modified source and evidence combination rule of evidence is put forward, and uses which as the fusion algorithm of decision makers, and a hierarchical structure of generator fault diagnosis model is set up, which is used to fuse the basic probability fuction separately obtained by two kinds of fault diagnosis on the layer basic to get the more reasonable and more accurate result of fault diagnosis.