Fault Diagnosis and Location of Power Transformer Based on Optimized Deep Belief Network with Random Forest and Sparrow Search Algorithm

Abstract

The running status of the power transformer has a crucial impact on the security and stability of the entire power system. The high-precision equipment diagnosis method is the most important part in judging the transformer condition. Consequently, a transformer fault diagnosis model based on optimized deep belief network with random forest and sparrow search algorithm (SSA-RF-DBN) is proposed. The deep belief network (DBN) is exploited to mine information from historical fault data of transformers and provide diagnosis results. The input and model parameters of DBN are optimized and determined by random forest (RF) and sparrow search algorithm (SSA) respectively. The multi-layer diagnosis structure including fault type diagnosis, electrical and thermal fault location diagnosis is established based on diagnosis models. The superiority of SSA-RF-DBN is verified by comparison experiments with other algorithms. And the diagnosis structure is applied on a transformer in service. The experiment results prove that the method of fault diagnosis and location diagnosis of power transformer proposed in this paper not only possesses better accuracy and reliability, but also makes up for the weakness of traditional methods in evaluating the fault location.