Supervised Machine Learning to Identify Loss of Excitation and Power Swing in a Generator

Loss of Excitation (LOE) and Power Swing (PS) are some of the significant conditions that might arise in a synchronous generator leading to the loss of power system stability. In this paper, supervised learning-based methods is proposed for early identification and classification of these conditions during occurrence of different type of faults. For this, Random Forest (RF) Classifier and Feed Forward Neural Network (FFNN) Classifier are used in this work. Different generator variables like active power, reactive power, load demand, terminal voltage, rotor angles are given as inputs to train these models. For fast and reliable detection of LOE and PS conditions, the performance of the train and test models is investigated on the data simulated on IEEE 39 bus system for different operating scenarios. The results highlight the suitability of the proposed method for fast and accurate online determination of these conditions in synchronous generator. The finding may be helpful for the system operators to take proper protective relaying decision in case of such conditions.