Classification of power quality disturbances using time/frequency domain features

This effort focus at identifying statistically significant time/frequency domain features with adequate separability and extensive interclass variability to discriminate probable Power Quality (PQ) disturbances such as sag, swell, harmonics, outage, transients etc. Understanding of disturbances is vital to investigate the origin and causes of PQ perturbances, events of equipment failure and for further classification as well as to employ mitigation measures. A database of PQ disturbances, simulated from their parametric equations, comprising normal signal, swell, sag, harmonics, hybrids of swell/sag and harmonics, outage and oscillatory transients was generated. Shannon entropy, global variance, maximum gradient, high/Low power spectral density ratio, skewness, kurtosis and percentage distortion were extracted from these simulated disturbed power signals. Optimum decision boundary was set and decision rules were framed, by examining the seperability and intra/inter class variability and range of the aforementioned statistical attributes in 3D feature space, prior to synthesizing decision tree for classification. Shannon entropy, global variance, maximum gradient, High/Low power spectral density ratio, skewness, kurtosis and percentage distortion exhibited amiable separability and inter/intra class variability, adequate enough to facilitate a simple and computationally feasible rule based characterization of PQ disturbances.