Rapid Response Strategy for Transformer Gas Relay Fault Diagnosis Assisted by the IDBO-VMD-SVM Combined Model

Abstract

Power transformers are core components of power grids, and their accurate fault diagnosis is critical for ensuring secure and stable grid operation. To address the limitations of conventional gas relays: inadequate extraction of dissolved gas analysis (DGA) features in transformer oil, poor model precision, and susceptibility to imbalanced datasets, this paper presents an IDBO-VMD-SVM combined model-assisted rapid response strategy for transformer gas relays. A multi-dimensional feature space consisting of gas concentrations, proportions and ratios is established and the dataset is divided. A local random affine shadow sampling algorithm is adopted to oversample minority fault samples, which balances category distribution and prevents noise introduction inherent to classic oversampling techniques. The intelligent IDBO-VMD-SVM auxiliary model is constructed to achieve fast identification and operation of transformer internal faults. Numerical examples demonstrate the validity and practicability of the proposed approach, offering an efficient technical solution for online transformer fault diagnosis.