Research On Application Of Transformer Winding Hot Spot Temperature Prediction And Abnormal Diagnosis Technology

As one of the important equipment in the substation,the power transformer will seriously damage the normal and stable operation of the power system once the abnormal operating state occurs.The life cycle and load function of the transformer depends on its own thermal characteristics,that is,whether the heat generated inside the transformer can be dissipated to the surrounding environment in time.The winding hot spot temperature is an important indicator for measuring the thermal characteristics of the transformer.If the hot spot temperature exceeds a certain threshold,it is easy to cause bubbles,which will accelerate the aging of the power transformer and lead to the deterioration of the winding insulation performance.Based on this,the thesis conducts related research on the prediction and diagnosis technology of transformer winding hot spot temperature,mainly including the following aspects:(1)During the working process of the transformer,the oil temperature of the top layer is usually monitored,and the hot spot temperature of the winding can be estimated by using the oil temperature of the top layer.Therefore,if the abnormal temperature of the transformer is detected in time before the winding temperature exceeds the safety threshold,it is necessary to predict the value of the oil temperature of the top layer.In this thesis,a top oil temperature prediction model is established based on the historical top oil temperature data of a certain area,and a research on the top oil temperature prediction of transformers based on EMD-IPSO-LSTM is proposed.Firstly,the historical oil temperature data is decomposed into several relatively stable intrinsic mode function components by using the empirical mode decomposition algorithm(EMD),and then the LSTM model is established for these components,and the improved PSO algorithm is introduced to obtain the LSTM through iterative optimization.The best parameters of the model are established,the model is established for training,and the modal function components are finally reconstructed to obtain the predicted value of the top oil temperature.(2)The collected infrared images of transformer windings are preprocessed and data enhanced.First,the original data set is expanded by vertical mirroring,Gaussian blur,image noise,and image rotation.Secondly,due to the influence of external factors such as light and temperature,the image quality of some images is reduced.Therefore,the quality of the image data needs to be related to the processing.Finally,according to the VOC 2007 data set standard format,the LabelImg tool is used to label and classify all the winding infrared images,and a standardized data set for the diagnosis of transformer winding temperature anomalies is established.The concentration includes four winding states,which are normal,mildly abnormal,moderately abnormal,and severely abnormal.(3)For the transformer winding infrared image data set,a lightweight YOLOv4-tiny model is trained to diagnose the abnormal temperature of the transformer winding hot spot.Secondly,in order to further improve the accuracy of the model,the original feature fusion network is improved and added respectively.Different attention mechanisms are tested and compared,and finally an algorithm model with higher accuracy than the original YOLOv4-tiny network model is obtained.In this thesis,the EMD-IPSO-LSTM combined model is used to predict the oil temperature of the top layer of the transformer,and combined with the winding infrared image diagnosis model,the early warning and real-time diagnosis of the transformer winding hot spot temperature have certain engineering reference value.