Signal Interference Suppression And Recognition Research On Partial Discharge

Insulation deterioration is the important reason of the electrical equipment malfunction.If the electrical equipment insulation deterioration can be detected in the early stage andtimely replaced, it will be effective in reducing accidents.Partial discharge detection that belongs to transient weak signal reflects the deteriorationsituation of the electrical insulation equipment, and electromagnetic interference of thedetection scene is usually serious, so the detection signal interference suppression is atechnical difficulty of partial discharge detection. At present partial discharge fault diagnosismainly depends on the threshold diagnosis, it will emit a warning signal when dischargereaches the minimum amount of hazardous discharge warning signal. But there are obviousdifferences in the extent of the damage that different insulation defects influence theinsulation system, the defect type recognition classification of partial discharge is a researchfocus on partial discharge detection.The main interference of partial discharge detection is white noise, the wavelet transformtechnology is usually applied to suppress white noise and its effect is more significant. Thisarticle has compared the wavelet de-noising algorithms and concluded that there are inherentdefects in the most commonly used de-noising threshold algorithm, the de-noising result isnot very satisfactory and stable enough. In view of this situation, the article has combined thewavelet threshold algorithm which adopts a novel threshold selection method and spatialcorrelation algorithm to jointly de-noise to the signal, and obtain the desired results.The most important part is the extraction of the characteristic quantities in the defect typerecognition of partial discharge. The fractal character is a characteristic parameter that has fewparameters and the mode distinguish ability. The article has conducted in-depth research inextracting the fractal characteristics of partial discharge scattered plot method, and madeimprovement to the traditional box dimension calculation method while extracting the fractalcharacteristics of scattered point sets.The fractal characteristics just describe scattered pointset of local self-similarity and complexity and not describe the scattered plot sets in thewhole, so the article has combined the gravity center of the scattered point set distributionand the fractal character to constitute the eigenvectors as recognition input vectors,and hascompared the recognition effect of the several eigenvectors in the test, the test results haveshown that the extracted eigenvector mode has distinguished better.BP neural network classifier is a commonly used defect type recognition classifier of partial discharge. But the traditional BP algorithm has inherent defects which restrict the useof the classifier in practice. The paper has applied the Particle Swarm (PSO) algorithm tooptimize the BP network. It can not achieve the desired result if the traditional PSO algorithmis applied to optimize the neural network in the recognition, So The article has taken theinertia weight PSO algorithm to optimize the network and applied in the defect typerecognition of partial discharge,the results has shown that the optimized network recognitionrate improves significantly.Finally, the content and results of the full text have been summarized, and deeplyresearch as well as some new ideals are proposed.