Analysis of Intelligent Classification and Prediction Model for Distribution Network Faults Based on Data Mining Structure

The rapid progress of smart grids has continuously increased the requirements for data balance processing and classification accuracy in fault classification and prediction of distribution networks.Given this, this study proposes a fault classification prediction method that combines clustering assisted generative networks and random cost weight convolutional networks.By using clustering assisted generation networks to balance and optimize the original data, the problem of inter class imbalance in distribution network fault data has been alleviated.Random cost weighted convolutional networks improve the classification and prediction performance of minority class fault samples through cost sensitive loss functions and deep feature extraction techniques of convolutional layers.In performance testing, the area under the curve of the imbalanced dataset processed by clustering assisted data generation method was 0.97, and the training time was 340.7 seconds.In the simulation test, the model classification prediction accuracy at 50 step size was 91.7 %, the average prediction delay was 20.3 ms, and the state transition error rate was 6.5 %.The results show that the proposed method performs well in classification accuracy and minority class recognition performance, and can effectively satisfy the classification demands of complex fault modes in distribution networks.This paper offers a new solution for the precision and intelligence of fault diagnosis in distribution networks, which has certain engineering application value and practical significance.