Fault Diagnosis Of Wind Power Transformer Based On Feature-weighted Neutrosophic C-means Algorithm

As a renewable clean energy,wind power has been vigorously developed in China in recent years.The scale of wind power generation has been continuously expanded,and the technology of wind power generation has become more and more mature,which still has a broad development prospect.When the wind is unstable,the output voltage of the wind power system will also be unstable,which will cause various faults in the wind power transformer,and even a major accident may occur.Therefore,accurately detecting the fault of the transformer and analyzing the typical fault at the initial stage of the fault is an important measure to ensure the safe and reliable operation of the wind power transformer.This paper focuses on the fault diagnosis of wind power transformers,and combines the intelligent learning algorithm with the neutrosophic theory.The main research and analysis are as follows:(1)The background and research significance of this paper are briefly described.The research status of transformer fault diagnosis and fuzzy C-means algorithm at home and abroad is analyzed.The common faults of transformer and the relationship between the characteristic gases when the faults occur are studied deeply.(2)Two basic fuzzy clustering algorithms HCM and FCM are introduced,and the neutrosophic C-means clustering algorithm(NCM)is further proposed.This algorithm introduces the redistribution of the distribution of samples by neutrosophic theory,and creatively adds uncertainty to the FCM.Uncertainty reflects the relationship between the sample and the neutral region by examining the relationship between the sample point and the neutrosophic point.The advantages anddisadvantages of the three algorithms are also detailed in this paper.(3)In the transformer insulation fault,the content of the main gas and gas components produced by different fault types is very different.The sample composed of the dissolved gas component content in the oil is characterized by different degrees of typical dimensions.The impact is also different.The general clustering method does not take into account the difference in weight between the various dimensions of the data.In this paper,by adding feature weights to the traditional FCM algorithm,each dimension of the sample dataset has primary and secondary divisions,thus highlighting some feature pairs.The main influence of clustering reduces the interference of redundant features.Through comparison experiments,this method can optimize the distribution of data and greatly improve the clustering accuracy.The feature weight matrix can reflect the different functions of the datasets in the clustering process.(4)Based on the above,this paper proposes a Feature-Weighted Neutrosophic C-Means Algorithm.The algorithm introduces feature weighting and neutrosophic theory on the classical FCM algorithm,which can redistribute the weight and distribution of samples,making the layout of some unbalanced samples more reasonable and the sample precision improved.For neutrosophic points,it has important practical significance.Through the analysis of the neutrosophic point generated by the neutrosophic division,it is possible to predict the next possible change of a certain type of fault,which has a guiding role for the subsequent development of the fault.It has important reference value for fault diagnosis of transformers,and the practical significance of faults can also be reasonably explained.