Multi-class support vector machines for static security assessment of power system

Abstract

Accurate determination of present operating state of power system takes utmost importance for utility operator for reliable and secure operation of power system. Conventional power flow methods have the limitations of larger memory requirement, longer computational time and infeasible options for real time applications for security assessment. Moreover, using performance index based on bus voltage deviations and line loadings for security assessment suffers with masking problem and fails to discriminate the contingency cases with the closer violations. Hence, composite security index has been formulated in this work addressing hyper-ellipse encompassed within a hyper-box concept to address multiclass classification problem of power system security using support vector machine (SVM). Sequential forward selection has been implemented for optimal feature selections to achieve the highest classification accuracy and the least mis-classification rate. The results of the proposed method have been validated and compared with existing method for three IEEE standard test systems.