Loss of Field Protection in Synchronous Generators Based on Data Mining Technique

Abstract

Loss of field (LOF) is a common fault in synchronous generators. Mason-Berdy scheme is the well-known and practical protective scheme to detect the LOF conditions. However, this impedance based method comes up with mal-operations under special situations such as under excitation operation, power swing phenomena, and partial LOF. In this paper, a new method based on data mining scheme is proposed considering the statistical correlation and zero-crossing function of electric variables as input training features. Moreover, new scenarios including special conditions such as condenser mode in both leading and lagging power factors are considered. In order to detect the LOF fault, the support vector machine (SVM) classifier is used. Results of the proposed method are comparable with conventional and improved Mason-Berdy schemes considering dependability, security and accuracy criteria.