Wind Turbine Fault Diagnosis Method Based On Information Fusion

With the development of the modern economy, in the last decade, as a representative of a new energy, clean energy, wind energy is in the rapid development of the wind farm at home and abroad when the accident occurred. The wind turbine itself bearing wear, gear break age the destruction of the wind turbine blade damage flying and other common failure will lead to a certain extent, caused by the occurrence of safety accidents. In view of these problems, the safe operation of the wind turbine and fault diagnosis has gradually become a research important for the wind power industry.Actual fault diagnosis due to the diversity of the complexity of the structure of large equipment and operating conditions as well as different operating environment, will introduce a lot of uncertainty, such as:diagnosis results based on different characteristics sometimes conflict; based on different locations the results of the diagnostic sensors sometimes conflict. These can lead to diagnostic accuracy and reasonableness of decline; it is difficult to meet the needs of increasingly large and complex equipment fault diagnosis. This article is in this context, the algorithm of information fusion technology research and application to fault diagnosis of wind turbine.The main research results are as follows:(1) This paper makes a detailed study of the content of information fusion algorithm D_S evidence theory algorithm, including the basic concepts of the algorithm (such as:basic probability assignment function, function of trust, like the true function and the reliability function, etc), the algorithm combination rules, the advantages and disadvantages of the algorithm, based on the improved methods of the algorithm, and the comparison between these improved methods, so as to have a comprehensive deep understanding of D_S evidence theory.(2) According to the traditional D_S evidence theory algorithm to evidence there is evidence to high conflict situation difficult to obtain accurate diagnosis problems, puts forward some suggestions for the improvement of evidence based on the modified source evidence. Is mainly to the conflict strength concept and the introduction of the concept of evidence mean distance, the use of the individual or the mean intensity of conflict evidence the concept of distance are cannot be effectively characterization between the degree of evidence. This paper is the product of both the geometric mean to represent the conflict between the evidence, and then use geometric mean of reciprocal to give the conflict evidence different weight, finally reuse D_S evidence combination rules of fusion, through the numerical examples validate the rationality and validity of the method.(3) According to the evidence theory algorithm of evidence from the basic probability assignment function is difficult to obtain, the paper adopted based on random set rough set-neural network method for obtaining basic probability assignment. Mainly is the first we adopt random set said the rough set attribute reduction of condition attribute reduction, reduce the neural network’s input dimension, and then the BP neural network training, and its output normalized get a evidence probability assignment. This paper, by using this method is used to obtain the probability of evidence after assignment of the improved algorithm to wind power generation units gear box fault diagnosis application, and finally got the reasonable diagnosis, and with other three methods comparative analysis, verify the effectiveness of the improved method, the practical application has a certain practical value.