Power grid stability identification using high discriminative factors

ABSTRACT The aim of this work is stability identification considering influence of different factors such as reaction time, nominal power and price elasticity of consumers and producers in stability situation of the power grid. The binary coding-based feature weighting (BCFW) method is introduced. The proposed method assigns greater weights to the factors with higher ability in separation between stable and unstable states in the classification process. To this end, the binary vectors of the first statistics of the data samples are generated. According to the defined binary vector for each factor (feature), that factor belongs to one of four possible states. While two possible states are appropriate for separation of stable from unstable situations, two other ones are inappropriate for this purpose. Feature weighting is done according to the defined states. The proposed method shows superior performance compared to support vector machine (SVM), multinomial logistic regression (MLR), convolutional neural network (CNN), maximum likelihood (ML) and nearest neighbour (NN), especially using limited training samples. With using just 1% training samples, the proposed BCFW method identifies the stability situation with 80.01% overall accuracy, while SVM, MLR, CNN, ML and NN achieve 79.41%, 79.05%, 71.91%, 71.24% and 64.04% overall accuracy, respectively.