As the electric power industry rapidly advances, diagnosing faults in electrical equipment has emerged as a critical challenge. In this study, we leverage advancements in power information technology to develop a method for extracting feature volumes, incorporating multiple characteristics to address mixed faults. Our approach begins with the application of a Backpropagation (BP) neural network to extract fault features from electrical equipment. Subsequently, we employ a Bayesian-optimized Correlation Vector Machine (CVM) classifier for precise diagnosis of mixed faults in transformer windings. We evaluated the performance of our BP-RVM model against traditional diagnostic models through real-world electrical fault diagnosis tests. The results demonstrated high diagnostic accuracy, with a maximum of 99.19%, a minimum of 96.41%, and an average accuracy of 97.73%, indicating a significant success rate in diagnosing types of electrical equipment faults. By providing new theoretical support and technical guidance, this study can improve the accuracy and efficiency of electrical equipment fault diagnosis.