Use of hidden Markov models for partial discharge pattern classification

An attempt was made to use hidden Markov models (HMM) to classify partial discharge (PD) image patterns. After an introduction to HMM, the methodology and algorithms for evolving them are explained. The selection of the model and training parameters and the results obtained are discussed. The utility of the approach is evaluated by applying it to five types of actual PD image patterns. The performance of the HMM approach is shown to exceed that of neural networks. >