Wavelet transform and artificial intelligence based condition monitoring for GIS

A condition monitoring (CM) system, which integrates wavelet analysis and artificial intelligent techniques to analyze the partial discharges (PD) patterns and classify the defective equipments in gas insulated switchgear (GIS), is presented here. The multi-resolution signal decomposition (MSD) attribute of the wavelet transform is employed to de-noise the PD signals measured on site, due to its spectrum decomposition capability. The performance of several well tried wavelets are investigated in the case of denoising, and this demonstrates that the combined use of the Biorthogonal and Daubechies type of wavelets achieves good de-noising results compared to other wavelet families. Different feature extraction methods are applied to the de-noised PD signals, to form pattern vectors which are further used as the inputs to a neural network based classifier, so as to identify the PD patterns and defective equipments in GIS. Several types of neural networks, both supervised and unsupervised (trained), are then evaluated with regard to their suitability as classifier according to training time and classification accuracy. Finally, the performance of the proposed CM system is ascertained by using a set of PD signals emanating from defects in a circuit breaker (CB), disconnect switch (DS), bus bar (BS) and insulation spacer (SP) in practical GISs in Korean 154 KV high voltage transmission networks.