The occurrence of mechanical faults in On-load tap changers (OLTC) is a significant factor contributing to power transformer failures. The analysis of motor current signals and OLTC vibration signals offers an effective means to determine the operational status of the equipment. The signals were processed and eigenvalues were extracted using wavelet transform (WT) and variational modal decomposition (VMD), while fault types were categorized using Optimized Support Vector Machines Based on the Gray Wolf Algorithm (GWO-SVM). This study proposes a neural network integrated decision model that combines the K-Nearest Neighbors (KNN), Hidden Markov Model (HMM), and GWO-SVM algorithms to establish a diagnostic model for OLTC faults. The identification accuracy rate of the proposed model exceeds that of conventional solutions, demonstrating its significant practicality in OLTC fault diagnosis.