Transformer Fault Diagnosis Method Based on SRBM-VPMCD

In response to the limited performance of traditional transformer fault diagnosis in small sample classification under massive monitoring data, this paper proposes a hybrid diagnostic framework that combines variable prediction model (VPMCD) and stack constrained Boltzmann machine (SRBM). Firstly, the original feature set is constructed based on transformer oil color spectrum data, and dimensional differences are eliminated through normalization preprocessing; Secondly, using SRBM to extract high-order abstract features of data layer by layer unsupervised pre training, optimizing the network topology and hyperparameters; Subsequently, a fault classifier was constructed using an interactive quadratic regression model based on VPMCD; Finally, a small number of labeled samples are introduced to fine tune the hybrid model end-to-end, achieving global parameter optimization. Experiments have shown that this method outperforms traditional methods in terms of fault recognition accuracy and model generalization ability, providing an efficient solution for transformer fault diagnosis under complex operating conditions.