Parallelized Partial Discharge Feature Fusion and Pattern Recognition Based on MaxCompute

With the rapid growth of the power grid scale and the increasing complexity of grid structures, the scale of partial discharge data in electrical equipment has gradually increased from the TB level to the PB level. The extraction of partial discharge characteristics is a crucial step in pattern recognition to accurately represent the features of partial discharge and improve the accuracy of identification results. In this paper, a parallel partial discharge feature fusion and pattern recognition method based on MaxCompute is proposed. Addressing the problem of information loss and the need to improve identification accuracy caused by single signal-level features or image-level features, this paper uses the ResNet50 residual neural network to extract PRPD spectrum features and fuses them with statistical features through Principal Component Analysis and Weighted Load Evaluation (PCA-WLE). The fused features are input into a Support Vector Machine (SVM) for classification and identification. Furthermore, t-Distributed Stochastic Neighbor Embedding (t-SNE) is used to compare and visualize feature distributions. The algorithm's computational speed is compared in three different computing environments. The results demonstrate that compared to single feature extraction methods, the proposed feature extraction method exhibits better clustering discernibility. When compared with common algorithms, the proposed method achieves over 95% accuracy in the identification of four types of partial discharge. In the MaxCompute environment, the algorithm maintains fast and stable computational performance, confirming the reliability and superiority of the proposed method.