Particle swarm optimization feature selection for the classification of conducting particles in transformer oil

The determination of particle type and dimensions in transformer oil is accomplished by using a Particle Swarm Optimization (PSO) technique in terms of the features extracted from the measured partial discharge (PD) pulse patterns. PSO selection of effective features is shown to be successful with intelligent classification for both electrical and acoustically measured data. Classification results of individual measurements were also reliable and far surpassed the efficiency of classification results obtained using the classifier solely for the same dimension of input features. The approach in this paper provides a solid basis for a data mining technique that can be used for the interpretation of both time and phase resolved raw PD patterns by searching a wide range of statistical attributes.