Abstract:
This paper proposes a new technique for power quality assessment, using rule-based S-Transform (ST) as a feature extraction tool and Adaptive Boost (AdaBoost) as a classifier. Since detection capability of ST highly depends on the nature of Gaussian window, a simple rule-base is created for selecting a suitable window under various nonstationary signal conditions. Keeping a reasonable time-frequency localization of the disturbance signals in view, the rule-base is prepared using statistical based entropy measure. In classification stage, decision stumps are used as weak classifiers and the strong classifier is constructed as a linear combination of weak classifiers in the AdaBoost algorithm. The performance of the proposed methodology is further improved using adaptable initial weights and a simple strategy to avoid overfitting of classifier. The efficacy of the proposed method is demonstrated using synthetic as well as Real Time Digital Simulator test data. The results reveal that the proposed rule-based ST and AdaBoost based method performs better than the other methods viz., SVM and Decision Tree (DT), under varied noise conditions as well as under varied amount of data used for training.
