Intelligent Diagnosis of Power System Secondary Equipment Based on Selective Bayesian Networks and XGBoost

Aiming at the problems of high-dimensional and unbalanced fault data of the secondary equipment of power systems, as well as the fact that the diagnostic model is prone to overfitting, which leads to a bias in the fault diagnostic results, we propose an intelligent diagnostic method for the secondary equipment of power systems based on selective Bayesian networks and XGBoost. The intelligent diagnosis of the power system secondary equipment is divided into four links: the signal preprocessing link adopts the time-domain feature extraction method to extract the feature vectors in the original power system secondary equipment fault signals, which are used for data enhancement and classification model training, and the signal generation link is based on the temporal generative adversarial network (GAN) model to train on a few samples of the fault class of the power system secondary equipment and to generate fault samples for balancing the power system secondary equipment fault samples. The training fault diagnosser uses the integrated learning algorithm XGBoost to build the fault diagnosis model, takes the balanced fault data set as the input of the fault diagnosis model, trains the fault diagnosis model, obtains the fault classifier, and effectively prevents the model from overfitting by introducing a regularization item and other technologies. Hyperparameters such as the number of classifier decision trees were optimized by selective Bayesian networks to improve the performance of intelligent diagnosis of the secondary equipment of the power system. The experimental results show that the data enhancement method of Time GAN is feasible for fault data generation. The parameter optimization process has good stability and is highly applicable in fault diagnosis.