Uncertain Information Fusion Method Based On Random Sets And Application On Fault Diagnosis Of Wind Turbine Generators

There are a lot of uncertain information processing methods. Fuzzy sets theory, vague sets theory and DS evidence theory are researched in this paper. Fuzzy set is that relation between set and elements is described by membership function. Vague set is that relationship between set and elements is described by true membership function and false membership function. The key idea of evidence theory is that the different evidence can fuse using evidence fusion formula. Fusing results is uniform description to objective with all evidence. There are two problems while making use of DS evidence theory. It is hard to get basic probability assignment (BPA). It is difficult to use evidence fusion formula, because of high conflict between BPA.Aiming these problems, the research finds and innovation of this paper are mainly indicated on two aspects as follows:(1). Basic concepts and theories of information fusion are introduced firstly, and uncertainty of fault diagnosis field is minutely analyzed. Fuzzy set and evidence theory based on random set are summarized. Vague set based on random set is researched, which is a innovation. Common fault types and traditional fault diagnosis method are analyzed. Feasibility analysis on application of wind turbine faults diagnosis which is based on uncertain information fusion ways of random set is researched.(2). To the problem of gaining BPA difficultly, two methods are advanced. BPA is get making use of gauss membership function when primary information of many sensors is collected. Making use of random set of vague set, fault types waiting and standard fault types are described as random sets, which can matched based on believed measurement and possibility measurement of random set. These two methods which can get BPA of evidence theory are proved efficiently by experiments.(3). Two methods are also proposed, aiming to the problem of evidence conflict. The first way is evidence revised based on evidence support matrix, considering distance function and conflict factors of evidence, building a evidence support matrix which has a max eigenvalue. Normalization eigenvector of the max eigenvalue can sever as weight of evidence. The second method is evidence fused formula improved. Defining generalized set-mapping, new evidence fused method can be built, using random set model of evidence and set-mapping.