Research on Abnormal Diagnosis Algorithm of Power Transformer Based on Generative Adversarial Network

Power transformers’ vibration signals such as acoustic patterns are particularly useful in the diagnosis of mechanical type defects such as loose core clamps or faulty windings. Due to the scarcity of anomalous signals from the transformers and the redundancy of normal vibration signals, we focused in our current study on four Generative Adversarial Network(GAN), based on the existing theory, the results are improved accordingly with the actual conditions of transformer operation. In this paper, we use vibration data collected in the field and generate corresponding fault data according to different types of anomalies as a way to test the noise immunity and sensitivity of different algorithms. GAN as a very advanced deep learning theory in recent years, we apply it to transformer anomaly detection and compare it with some traditional methods of machine learning, and the final experimental results show that, deep learning anomaly detection algorithms generally yield higher accuracy rates, and each algorithm works better on its own anomaly detection scenario.