Integrating canonical correlation analysis with machine learning for power quality disturbance classification

Recently, the rapid growth of Renewable Energy Resources(RER)in power generation has resulted in the frequent occurrence of Power Quality Disturbances(PQDs)within the power system. The timely and accurate detection of these PQDs is critical for maintaining good power quality while integrating RER into hybrid power systems to make them more robust and stable. In this paper, a multi-view dimensionality reduction approach based on Canonical Correlation Analysis(CCA) is proposed to differentiate different types of PQDs. Here, a dataset of 29 types of PQDs which include nine single types and twenty multiple types of PQDs have been generated using their mathematical model in MATLAB for experimentation. CCA being multi-view dimensionality reduction technique maximizes the correlation between two different views of the data. Here two cases of datasets have been considered for further exploration, Case 1: PQDs without noise and with 20 dB noise, Case 2: PQDs with 20 dB and 30 dB noise. Furthermore, to test the efficacy of CCA in both cases, the extracted features have been tested using four different classifiers i.e. K-Nearest Neighbour(KNN), Support Vector Machine (SVM), Naive Bayes(NB), and Random Forest (RF). The performance of each of the classifiers has been tested on five different performance metrics such as precision, recall, F1 score, hamming loss and accuracy and the results shows that the proposed technique of multi-view dimensionality reduction is capable of classifying the PQDs with two different views at a time