Notes

conclusion
In this report, a different PQ monitoring approach based on EWT, HHT, and Multi SVM was proposed. EWT was used for signal conditioning technique and two distinctive features, IA and IF, were obtained by HHT. Then a series of statistical calculations was performed to form feature matrix. The algorithm is found capable of classifying many PQ disturbances with a high accuracy. SVM with polynomial (g= 0.95,d=2) kernel function was used for the decision-making unit and presented high accuracy among the other classifiers. The proposed algorithm was applied to the data-set generated on a computer .Proposed hybrid algorithm can be used in power quality monitoring software. The analysis and results presented in this work clearly show the potential capability of the suggested hybrid algorithm in classifying the distorted PQ waveforms.

results
To evaluate the performance of proposed method for the classification of PQ disturbances, 7 single and 7 combined disturbances have been considered.
The classes include Interrupt(C1), Sag(C2), Swell(C3), Harmonics(C4), Flicker(C5), Oscillatory Transient(C6), Notch(C7), Spike(C8), Sag combined with Harmonic(C9), Swell combined with Harmonic(C10),Sag combined with Oscillatory Transient(C11), Swell combined with Oscillatory Transient(C12), Flicker combined Harmonics(C13), Oscillatory Transient combined with Flicker(C14),table[]. 200 signals are generated for each class. From these 200 signals 150 are used to train the SVM and 50 are used for testing purpose. Fundamental frequency of 50 Hz has been chosen and the experimental signals are sampled at 10KHz. Window length of .2 sec has been selected for better frequency resolution and for better harmonic and inter harmonic estimation. Fundamental frequency variation of 0.5Hz, phase variation of 0 to 180, SNR variation of 25 to 55 and intensity variation of the disturbance is considered for the generation of experimental signals. LIBSVM which is an efficient tool for the SVM related training is used to evaluate the classification performance of extracted feature vectors for PQ disturbance signals
SVM is trained with linear, Polynomial, RBF and Sigmoid kernel function and the accuracies are shown in the table []. Polynomial gives better results in comparison of other Kernels. From the table it is clear that classifiers are able to classify single disturbances more precisely than combined disturbances.
14X14 Confusion matrix [table ()] is constructed to show the classification performance of the method. The diagonal elements represent the correctly classified class and the off-diagonal elements represent the misclassification. As shown in the table, Class C3 (swell) is misclassified with C12 (Oscillatory transient combined with swell), C8 (Spike) is misclassified with C6(Oscillatory Transient) due to the presence of similar characteristics. Few signals of combined disturbances are misclassified with one of its single disturbances due to the dominating behavior of that disturbance.
A performance comparison of noise less and noisy signals has been shown in [table()] the proposed technique gives better results for noisy signals in comparison of other techniques.