Reinforced integrated deep transfer learning network based on multi-fault domain transformer fault voiceprint recognition

Power transformers emit continuous vibration signals during operation. The signals contain a large number of pulses and fluctuations caused by mechanical faults. They are the main data source for evaluating the operating status of power transformers. In order to study the voiceprint characteristics of vibration acoustic signals of transformers under different operating conditions, a reinforced integrated deep transfer learning network (REDTLN) for multi-fault domain voiceprint recognition is proposed. First, use various new kernel maximum mean difference (kMMD) to build multiple deep transfer learning networks (DTLN) for individual origin single destination field adjustment, so that Make DTLN transferable features diversified. Additionally, a novel unified unsupervised ensemble learning is formulated according to maximum mean difference (MMD) and various metrics, and using this unified measure as motivation, A novel training approach is constructed to search for Efficacious combinations of these DTLNs Rules to achieve enhanced integration of multi-source deep transfer learning networks. Improved accuracy and robustness in dynamic noise environments and complex operating conditions. Empirical outcomes indicate that the suggested model can accurately determine the abnormal operating status of the transformer and diagnose six typical faults of the transformer in different noise environments, with a diagnosis accuracy of more than 99%.