Research On Transformer Fault Prediction Method Based On Ensemble Learning

As an important equipment of the power system,the power transformer’s operating status is directly related to the production and operation of the power grid.A considerable part of the oil-immersed transformers put into use in our country have been working for decades under the influence of electric field,thermal and mechanical fields.There is a risk of internal failure,and the internal failure of the transformer has a certain incubation period,and there is room for prediction.Dissolved gas analysis(DGA)technology is widely used in the field of transformer fault diagnosis.However,most current fault diagnosis techniques can usually effectively classify data close to different faults and have poor accuracy Based on the above analysis,combined with the advantages of ensemble learning algorithm in reinforcement learning effect,a transformer fault prediction method based on ensemble learning algorithm is proposed in this paper,which realizes the short-term prediction of gas content in transformer oil,and classifies the predicted gas fault,so as to achieve the purpose of fault prediction.At the same time,this paper optimizes data processing,feature selection and model construction to improve the prediction effect.This article mainly conducts research from the following aspects:First,taking the dissolved gas concentration in transformer oil as the prediction object,a VMD-XGBoost gas concentration prediction model is proposed.According to the gas history sequence information,VMD is used to decompose the gas concentration sequence.The historical data of each sub-sequence is used as the input of the XGBoost prediction model.The forward verification method is used for short-term prediction,and then the time step and model hyperparameters are searched.It improves the accuracy of the gas concentration prediction model and provides reliable input data for the fault diagnosis model.Secondly,in view of the low accuracy of the traditional transformer fault diagnosis model and the fuzzy selection criteria of fault characteristics,a transformer fault diagnosis method based on LightGBM is proposed.With reference to multiple feature selection methods,combined with two feature importance screening methods,a combination of features that can better reflect the different operating conditions of the transformer can be extracted from the fault data set.The evolution process of binary classification problems to multi-classification problems is described,a fault diagnosis model based on LightGBM is established for model training,and the hyperparameters are adjusted through the grid method to improve the accuracy of diagnosis.Using multiple evaluation indicators to compare the diagnosis results of multiple fault diagnosis methods proves that the model proposed in this paper has better performanceFinally,combined with the research results of the first two parts,the actual failure cases are predicted and analyzed.Based on historical data,predict the gas concentration within a certain period of time before the fault warning,and use the fault diagnosis method to classify the prediction results.The classification results are compared with the online monitoring and diagnosis results and the actual oil test results,which proves that the fault prediction method proposed in this paper can accurately predict the actual internal faults of the transformer in advance.