Edinburgh Napier University

  This paper presents an ensemble-boosting algorithm (EBA) for classifying partial discharge (PD) patterns in the condition monitoring of insulation diagnosis applied for electrical assets. This approach presents an optimization technique for creating a sequence of artificial neural network (ANNs), where the training data for each constituent of the sequence is selected based on the performance of previous ANNs. Four different PD faults scenarios were manufactured in the high-voltage (HV) laboratory to simulate the PD faults of cylindrical voids in methacrylate, point-air-plane configuration, ceramic bushing with contaminated surface and a transformer affected by the internal PD. A PD dataset was collected, pre-processed and prepared for its use in the improved boosting algorithm using statistical techniques. In this paper, the EBA is extensively compared with the widely used single artificial neural network (SNN). Results show that the proposed approach can effectively improve the generalization capability of the PD patterns. The application of the proposed technique for both online and offline practical PD recognition is examined.