Machine Learning Approach to Differentiate Excitation Failure in Synchronous Generators from Power Swing

Abstract

Loss of Excitation (LOE) is the most considerable fault in Synchronous generators since it affects both the generators and power network. The traditional protection method for LOE is based on impedance trajectory of the machine with negative offset mho relay. Meanwhile the traditional method experiences malfunctions and speed dip in LOE detection. This paper presents machine learning approach to detect LOE fault as well as classification logic to discriminate LOE fault from power swing conditions due to Line fault. This paper utilizes Hotelling’s-T2 statistical method to calculate Hotelling’s-T2 based Fault Indices (HT2 -FI) for fault detection and Support Vector Machine (SVM) for classification. The time series data of electrical quantities such as Terminal voltage and Reactive Power of the generator are extracted from simulated Single Machine Infinite Bus test system and used as input data. This data is involved in calculation of HT2 –FI and in development of classification logic. The proposed method is simulated and verified for complete, partial LOE conditions and power swing conditions. Simulation outcomes depict the remarkable signs of the proposed method in LOE identification from power swing. Comparative assessment also reports that the method is capable of saving time in detecting LOE.