Online DGA Sensor Calibration Using GANN and Data Augmentation

The online dissolved gas analysis (DGA) sensors can diagnose the incipient faults of the transformer in real-time. However, these sensors suffer from low accuracy, and may provide unreliable even misleading maintenance suggestions. To tackle this problem, this paper proposes a hybrid genetic algorithm neural network (GANN) to adaptively build the calibration model for online DGA devices. As offline data is too few, a perturbation data augmentation method is also proposed to generate synthetic offline DGA data. The online data can be calibrated by the model thus providing real-time power transformer condition monitoring. The comparison with popular calibration methods confirms the superior of the proposed method.