MCSVM and MCRVM based Contingency Classification Model

In this study, a machine learning model for contingency classification of an energy system was developed. A greater cost reduction can be achieved from the huge amount of data extracted from the energy infrastructure, and analysis of early contingency detection is performed. The complexity of contingency analysis can be reduced by mining, which reduces the hardware use. In this research, two different machine learning algorithms, a Multi-Class Support Vector Machine (MCSVM) and a Multi-Class Relevance Vector Machine (MCRVM), are used to classify the different contingency levels using the mined data, which comprises the voltage, power generated, and angles from line. Combined data cleaning and analysis served as a data transformation technique. Principal Component Analysis (PCA) is used to reduce the dimensionality of data for classification. The trained model using Multi-class SVM and RVM was generated using the line data output mapped on the composite contingency index obtained from it. The model thus generated would act as a classification black box that would classify the condition as normal, alarming or lowly contingent. A MATLAB simulation was carried out on an IEEE 30 bus system and classification of the contingency into three levels of contingency was observed to be satisfactory.