Research On Fault Diagnosis And Intelligent Warning Of Transformer Based On Data Driven

With the continuous growth of the national economy,the growing demand for energy puts forward higher requirements for the security and stability of the power grid.Transformer is an important electrical equipment in power system equipment.Transformer fault prediction and intelligent early warning can reduce the probability of a transformer fault,which is the basis of ensuring the normal operation and maintenance of the system.When transformer common faults such as overheating,partial discharge,spark discharge and arc discharge occur,a variety of gases will be generated.Dissolved Gas-in-oil Analysis(DGA)is an effective method to detect transformer faults.Depending on its data results,combined with traditional methods such as IEEE and IEC guidelines,the fault status and types of transformers can be diagnosed.The accumulation of experience of traditional fault diagnosis and intelligent early warning methods,including the three-ratio method,plays an important role in the research of this subject.With the rapid development of artificial intelligence technology and the comprehensive coverage of transformer data acquisition,more and more machine learning algorithms play an important role in transformer fault diagnosis.In view of the scarcity of single transformer fault data and the complexity of common artificial intelligence methods,and the difficulty of combining traditional methods,this paper proposes a data-driven bagging decision tree transformer fault intelligent early warning method: collecting different transformer equipment fault databases to establish early warning databases,and making part of the historical detection data of the fault equipment before the fault occurs.Abnormal or abnormal trend data are labeled as early warning signal data for training.Bagging decision tree is used to learn a database and generate a judgment system,assisted by feature selection compromise and system parameter optimization.When the system judges the abnormal real-time oil chromatographic data,alarm will be issued immediately,and decision tree and traditional three-ratio method will be used to provide the basis for system decision-making.At the same time,aiming at the monitoring of the proposed model,an improved prediction model of monitoring quantity based on NARX neural network is proposed,which can predict the future value of gas analysis in oil in real time.The proposed fault warning and diagnosis model can further guarantee the future safety and stability of the transformer system by synchronizing all the predicted values.The scheme enlarges the scale of fault database,gives full play to the advantages of artificial intelligence algorithm,learns the rules of some abnormal or abnormal trend data in the historical detection data of fault equipment before the fault occurs,enlarges the scope of early warning and warns potential faults.The trained model can be converted into a logical judgment statement,which is less difficult to implement than other AI algorithms without the support of the algorithm library,and can be easily combined with traditional methods including three-ratio method.Based on the real fault data test of substation,the accuracy and fault diagnosis rate of the proposed scheme are higher than those of other common models such as neural network,K-nearest neighbor(KNN),support vector machine(SVM),linear discriminant analysis and logistic regression analysis.The proposed prediction model of monitoring quantity is better than the traditional NARX neural network model and other prediction models such as the support vector machine(SVM)and regression tree.The error of linear regression and Gauss process regression model is smaller,and the combination of them can improve the stability of the transformer system.