Research on transformer transverse fault diagnosis based on optimized LightGBM model

At present, the research on transformer vibration characteristics mainly focuses on a certain voltage level, which makes the diagnostic method applicable to a narrow range. In order to improve the universality of the diagnosis method, the vibration data of transformer with different voltage levels are collected and the calculation formulas of two important characteristic values are improved. In order to reduce the influence of data imbalance on model training, Deep Convolutional Generation Adversarial Network (DCGAN) is used to extend the data. In this paper, Light Gradient Boosting Machine (LightGBM) was used to build transformer fault classification model and combined with Bayesian optimization algorithm (BOA), which greatly improved the final classification effect. The results show that the improved features can increase the accuracy of diagnosis results by 51.5%, and the accuracy of LightGBM diagnosis model after Bayesian optimization can reach 98.9%, which can realize the horizontal fault diagnosis of multi-grade transformers.