Research On Fault Diagnosis Method Of Oil-Immersed Power Transformer Based On Machine Learning

Power transformers are indispensable components in power systems,and their safe operation is crucial to ensuring the stable operation of the entire power grid.The method of dissolved gas analysis(DGA)in oil is the most commonly used and effective method for internal fault diagnosis of oil immersed power transformers.The traditional offline detection method of DGA requires regular on-site power outage inspection,and the monitoring cycle is long.Nowadays,online transformer monitoring systems in power grid data centers can obtain a large amount of DGA data without power outage.Most of the existing transformer fault diagnosis methods have problems such as low diagnostic accuracy and inability to effectively apply to DGA online monitoring systems.Therefore,it is necessary to explore a fault identification method for oil immersed power transformers that has higher accuracy and can be applied to actual DGA online monitoring systems.Based on this,this thesis proposes an online monitoring and fault diagnosis method for oil immersed power transformers using DGA based on machine learning,which provides a reference for determining transformer status.The main research contents and conclusions are as follows:A new method for transformer fault diagnosis is proposed by combining PCA and RF algorithms.Firstly,by increasing the ratio dimension of the dataset,the feature dimension is enhanced,thereby enriching the sample information;Secondly,the PCA dimensionality reduction algorithm is used to remove redundant information generated during feature dimensioning,thereby better extracting the original data,and using RF models to classify its faults.Experiments have proven that combining PCA and RF algorithms can significantly improve the accuracy of transformer fault diagnosis,effectively verifying the feasibility and effectiveness of the combined model in transformer oil chromatogram data fault diagnosis.According to the fact that the actual DGA online monitoring system is not a small sample size fault diagnosis,but a large sample dataset containing more normal operation data,a 2D-CNN based first level fault diagnosis model for DGA online monitoring is designed based on the data type of power transformers.The experimental results show that the 2D-CNN model classifier has the highest diagnostic accuracy for actual DGA online monitoring systems,indicating that using this 2D-CNN model is beneficial to improving the accuracy of fault diagnosis.On this basis,this thesis designs a two-level diagnostic model based on the actual operation of the DGA online monitoring system,which combines 2D-CNN and PCA-RF,to better apply to transformer fault diagnosis.The experiment shows that compared to other single algorithm discrimination models and other combined discrimination models,the combined model using 2D-CNN model as the first level and RCA-RF as the second level has a higher accuracy rate for fault discrimination for actual DGA online monitoring data.Therefore,a two-level diagnostic model combining 2D-CNN and PCA-RF can be applied to actual DAG online monitoring systems.