Early fault diagnosis of transformer windings based on the improved MVMD-ELM

Aiming at the problems of weak early fault characteristics of transformer windings, large noise interference and insufficient accuracy of traditional diagnostic methods, this paper proposes an early fault diagnosis method for transformer windings based on improved multivariable mode decomposition and optimized Extreme Learning Machine (ELM). Firstly, taking the leakage magnetic field as the fault characteristic state quantity, the decomposition parameters are adaptively adjusted through Multivariate Variational Mode Decomposition (MVMD) combined with the Dream Optimization Algorithm (DOA), and the wavelet threshold method is combined to efficiently denoise the noisy signal and improve the signal quality. Secondly, multi-dimensional fault features such as correlation coefficient, asymmetry degree, distribution difference degree and Hausdorff distance are extracted to construct the DOA-ELM diagnostic model. The relevant parameters of ELM are optimized by using DOA to improve the classification performance of the model. The simulation and dynamic model experiment results show that the proposed method can effectively identify early faults such as axial compression deformation of windings and inter-turn short circuits. The diagnostic accuracy rates reach 98.33% and 96.67% respectively. Compared with the traditional Back Propagation Neural Network (BPNN) and Support Vector Machine (SVM) methods, it has effectively improved in classification accuracy and computational efficiency. This method provides an effective solution for the precise diagnosis of early faults in transformer windings and has high engineering application value.