Research On Multi-source Image Information Recognition Method For Grid Equipment Fault Diagnosis

The core of reliable operation of the power grid lies in the stable operation of power equipment.The rapid and accurate diagnosis of power equipment faults is one of the key links for the stable operation of power equipment.Due to the complexity of the operating environment of modern power equipment,the failure of power equipment is characterized by randomness and ambiguity.The traditional fault diagnosis method only combines a single fault feature and a pattern recognition algorithm,but only finds the optimal fault classification from a single angle,which is bound to have one-sidedness and affect the reliability of the diagnosis result.Facing the massive image information in modern power maintenance,this paper studies the fault diagnosis method of power equipment based on multi-source image information.Firstly,the fault types and characteristics of power equipment represented by insulators and arresters are analyzed.Aiming at the multi-source image information set consisting of infrared image and leakage current waveform image in power maintenance,combining with information fusion technology,an information fusion framework based on feature layer and decision layer is designed,and a fault diagnosis model of power equipment based on BP neural network(primary diagnosis)and D-S evidence theory(decision fusion diagnosis)is established.Then based on the multi-source image preprocessing,the extraction method of insulator temperature characteristics,texture features and leakage current time-frequency domain characteristics is studied.The effective feature quantities of the insulator fault states under different class characteristics are given.The relationship between time-frequency characteristics of leakage current and insulator contamination faults is mainly studied,and the eigenvectors representing insulator failure state are constructed based on the effective eigenvalues under these three characteristics.In the primary diagnosis,the traditional BP neural network is easy to fall into the local minimum solution.And the problem of slow learning rate,this paper adopts the additional dynamic parameter method and the improved algorithm of adaptive learning rate to improve the BP neural network.Based on this,the BP neural network structure of power equipment fault diagnosis is designed and the fault characteristics are realized.The purpose of input decision output is the basis for the next multi-decision fusion diagnosis.In the decision fusion diagnosis,a decision fusion diagnosis method based on D-S evidence theory is proposed for the evidence conflict in decision fusion.By retaining similar evidence,the method of modifying the basic probability assignment of conflict evidence reduces the impact of evidence conflict problem on decision fusion results,and uses Dempester combination rule to diagnose fault diagnosis of each evidence.Finally,this paper applies the improved BP neural network and improved D-S evidence theory to the fault diagnosis of insulators and arresters.The experimental results show that the method can effectively diagnose the contamination,moisture and normal state of the insulator,and also achieve good results in the diagnosis of lightning arrester deterioration and moisture,which fully verifies the feasibility and high efficiency of the method in power equipment fault diagnosis.