Transformer Anomaly Detection Using Deep Belief Network with Dynamic Adaptive Knowledge Transfer and Interval Type-2 Fuzzy Set for Uncertain Parameters

Power transformers are key and expensive components in power systems. Precise evaluation of the transformer state is critically essential. This article presents a transformer fault diagnosis technique utilizing a deep belief network (DBN) enhanced with dynamic adaptive knowledge transfer (DAKT) and interval type 2 uncertain parameter fuzzy set (IT2F) structure. Firstly, in accordance with the compaction aspects of chaotic phase space and the Lyapunov index in chaos theory as quantitative judgment, the optimal analysis data selection method is obtained, which reduces invalid data and improves calculation accuracy. Subsequently, a neuro-symbolic framework is introduced to elucidate the operational dynamics of the profound architecture, and a novel sparse deep belief network leveraging adaptive knowledge transition is constructed utilizing the knowledge integration technique of rule amalgamation and elimination. This network possesses a more sparse topology and superior learning efficacy. Secondly, a new enhanced deep learning algorithm-interval type 2 uncertain parameter fuzzy set is introduced the parameters and offsets of the deep belief network, which can better handle the uncertainty between the input data and model parameters. Strengthen the robustness and generalization capability of the model. Comparative analysis with shallow machine learning methods and deep learning methods shows that the average recognition rate of the proposed DAKT-IT2F-DBN method is $97.43 \%$, with good robustness and performance exceeding other fault diagnosis methods.