Research On Transformer Partial Discharge Pattern Recognition And Fault Location Method Based On Ultrasonic Signa

The phenomenon of partial discharge is a classic external manifestation of transformer insulation damage and deterioration,and the realisation of transformer partial discharge treatment is necessary and important to maintain safe system operation and to detect and respond to potential insulation accidents.Ultrasonic detection is an effective method for local discharge detection,so this paper uses ultrasonic signals as the basic information source to explore the transformer partial discharge problem from three aspects: signal noise reduction,pattern recognition and fault location.1.A wavelet threshold local discharge signal noise suppression method based on improved threshold estimation and improved threshold function is proposed.Firstly,a comprehensive entropy threshold estimation model is established by fusing the sample entropy information of the noise sequence and the noise reduction sequence,while a dichotomous variable step length non-linear threshold search method is proposed.Then,an improved threshold function with continuous derivability at full threshold between soft and hard threshold functions is proposed,and the function parameter selection method is also discussed.Experiments show that the proposed method of local discharge signal noise suppression can effectively obtain the optimal threshold estimate with sufficient noise reduction and good performance of oscillation suppression.2.A partial discharge pattern recognition method based on enhanced fusion distribution spectrum and transfer residual network is proposed.Firstly,the continuous wavelet transform spectrum and the short-time Fourier variation spectrum of the signal are fused,and the image texture is enhanced by data cleaning to propose an enhanced multispectral fusion of the map source method.Then,a transfer learning method is introduced to quickly achieve autonomous extraction of local discharge fault features and local discharge pattern recognition by fine-tuning the Res Net50 model.Experiments show that the proposed local discharge pattern recognition method can provide rich and high-quality feature information,while adapting the deep network model to limited sample size learning and obtaining good local discharge pattern recognition results.3.A partial discharge fault localization method based on ultrasonic boda time difference immunity estimation and heuristic search is proposed.First,a time delay estimation method based on non-linearized signal high-order cumulative correlation is proposed by integrating the processing methods of low-order,second-order and high-order quantities of signals.Then,a non-linear minimum optimized fitness function is established based on the localization model,and an adaptive inertia weighting factor is introduced to improve the crowd search algorithm,through which the local discharge coordinates are searched.The experiments show that the proposed local discharge fault localisation method has good time delay estimation accuracy in a multi-noise environment,accelerates the search convergence speed and obtains effective local discharge localisation coordinates on the basis of ensuring the quality of the localisation solution.