Transformer Fault Diagnosis Method Based on Data Feature Expansion and GSA-SVM

Aiming at the status quo that transformer fault diagnosis has low precision and poor accuracy, and manual fault detection is still used in production, this paper proposes a transformer fault diagnosis method based on data feature extension and Gravity Search Algorithm (GSA) improved Support Vector Machine (SVM). First, the features are extended to 20 dimensions using the feature ratio method. Second, the features are optimized using the random forest (RF) algorithm to remove redundant features, reduce data complexity, and improve the accuracy and prediction speed of the prediction results. Finally, the improved support vector machine model (GSA-SVM) using the gravitational search algorithm is used to optimize the prediction model and predict the data, and its prediction result has an F1 score of 88.2%, which is more accurate than the models of several other algorithms. The results show that the method can diagnose faults quickly and accurately, and has some practical significance in the field of transformer fault diagnosis.