Optimal feature selection using modified cuckoo search for classification of power quality disturbances

The widespread usages of sensitive equipment such as computers, controllers and microelectronic devices have placed immense burden on the grid operators to deliver high quality electrical power to their customers. To achieve this end, the power quality disturbances (PQD) within the network need to be minimized. In this paper, a method to enhance the performance of the multiclass support vector machine (MSVM) classifier using the modified cuckoo search (MCS) is proposed. The wavelet packet transform is used to extract the crucial features from the PQD waveforms; these features are utilized as the input data to the classifier. In order to achieve high accuracy, robustness and speed, the MCS optimizes the number of selected features, as well as the penalty factor and slack variable of the MSVM. The proposed combinatorial algorithm (MCS-MSVM) is tested using 31 categories of PQD events; the hypothetical data for these events are generated by the IEEE 1159 Standard parametric equations. From simulation, the classification accuracy is recorded to be 100% under the no-noise condition, while at the signal-to-noise ratios (SNR) of 10, 20, 30 and 40 dB, the accuracies are 98.40, 98.54, 99.14 and 99.64%, respectively. Moreover, the comparative assessment concludes that the proposed method is superior to other heuristics-MSVM classification methods, namely the GA, PSO, differential evolution, harmony search and the conventional cuckoo search. The practical performance of the MCS-MSVM classifier is validated using real-time PQD data of a typical 11-kV underground distribution network, obtained from a particular electrical utility operator. For benchmarking, comparisons are made to 17 most recent PQD classification techniques published in literature. It is found that the proposed method exhibits the highest accuracies and the lowest computation times under ideal and noisy environments.