Transformer occupies a key position in the power system. Its safe and stable operation is directly related to the reliability of the entire power grid. The traditional transformer fault diagnosis method has limitations such as low sensitivity and a relatively high misjudgment rate. With the development of machine learning technology

the fault diagnosis method based on intelligent algorithms has become a research hotspot. Among them

the transformer fault diagnosis method based on ISO-LSSVM has attracted much attention because of its good generalization ability and nonlinear processing ability. This paper first analyzes the common fault types of the transformer (such as overheating fault

discharge fault

insulation moisture

mechanical failure)

and extracts the fault characteristics such as dissolved gas analysis

electrical test parameters

temperature parameters

etc. in the oil

and preprocesses the characteristic data such as normalization

missing values and abnormal values. Then

a fault diagnosis model based on ISO-LSSVM is built

anomaly detection and feature selection are used for isolated forests

and fault classification is carried out through LSSVM. The model design ideas

parameter selection and optimization

and training and verification process are expounded. Finally

through the experimental verification of 1000 sets of transformer fault data of a power company

it is divided into 700 sets of training sets and 300 sets of verification sets. Compared with traditional SVM and BP neural networks

the results show that the accuracy rate of this method is 92%

the recall rate is 90%

and the F1 value is 91%. It has a high diagnostic accuracy rate and good generalization ability

which provides a guarantee for the safe and stable operation of the power system. In the future

more fault characteristics and optimization algorithms can be combined to further improve the performance.