Wind Turbine Fault Diagnosis Of Information Fusion Method Based On Evidence Theory

Wind power energy as the main representative of the renewable energy，plays an important part in current grid. It holds a hot spot in the windpower about troubleshooting effectively after the fault being occurred andthen removes it timely. Evidence theory indicates a strong advantage in theexpression and measurement of uncertain information as well as the fusionprocess，which has been applied in the fault detection and diagnosis oflarge equipment. This paper carried out the research with fault diagnosis ofwind turbine applying multi-source information fusion method based onevidence theory, then the main work of this paper including：（1）The basic concepts and main principles of information fusiontechnology are firstly introduced, and then carry out a detailed analysis ofits application in fault diagnosis. The basic framework and thecombination rule of evidence theory are studied and the fault diagnosismethods based on evidence theory are summarized. Analyzing the currentcommon method of fault diagnosis of wind turbine, then the feasibilityanalysis of multi-source information fusion method based on evidencetheory is made.（2）For the high conflict appearing in the fusion process of theoriginal evidence, a modified method based on the weight of evidenceentropy is proposed. Depending on the importance of evidence obtained bymultiple sensors, the evidence entropy can be applied to obtain the weightof each parameter. The framework of fault information fusion method isreached based on evidence entropy and well applied in the generatorbearings. The fault vibration signal extracted by bearing acceleration sensor is decomposed to obtain intrinsic mode function, and then take thecharacteristic frequency of envelope spectrum as fault indicator. Comparethe characteristic term with fault frequency to get the preliminary diagnosis.The original evidence is obtained by the gray correlation principle whichthrough weight adjustment by evidence entropy and then integrate withDempster rule.（3）In practical application of fault diagnosis with evidence theory,for the environmental impact and the limitation of sensors, there are stillsome drawbacks in extracting BPA, and then this paper presents a methodapplying clustering centralization. Take wind turbine generator as aresearch object; firstly clear its fault domain to determine the faultcharacteristic quantity. Mining the relationship between fault type and faultfeature, and then take the reciprocal value of Euclidean distance betweencenter of base point and the diagnostic mode as support degree, which hasthe same physical meaning with BPA. By utilizing the accumulatedinformation and current information effectively, the temporal-spatial fusionmethod can reach a comprehensive description of the fault and to someextent eliminates the redundant information to improve the accuracy ofdiagnosis.