Transformer Dga Fault Diagnosis Method Based on Data Reconstruction Combined With Idbo-Lssvm Model

Currently, the combination of big data mining techniques and dissolved gas analysis (DGA) monitoring can be used to diagnose transformer faults. However, due to the high dimensionality and imbalanced fault samples in DGA data in practical engineering, existing diagnostic methods generally suffer from low accuracy and poor fault tolerance. Therefore, this paper proposes a transformer DGA fault diagnosis method based on data reconstruction combined with the Improved Dung Beetle Optimizer-Least Squares Support Vector Machine (IDBOLSSVM) model. This method first employs the Adaptive Synthetic Sampling (ADASYN) method to synthesize minority class samples and utilizes t-Distributed Stochastic Neighbor Embedding (t-SNE) for feature extraction to reconstruct the DGA raw data samples. Subsequently, a transformer fault diagnosis model is constructed based on Least Squares Support Vector Machine (LSSVM), and the Improved Dung Beetle Optimizer (IDBO) is used to optimize the model parameters. Efficient acquisition of the global optimal solution is achieved by introducing the Sine Power Mapping (SPM) chaos map, adaptive weights, and Cauchy mutation strategy. Finally, the accuracy and superiority of the proposed transformer fault diagnosis method are verified through algorithm case analysis and comparative calculations with multiple methods using measured data.