Performance Analysis of Regression and Artificial Neural Network Schemes for Dynamic Model Reduction of Power Systems

Abstract

The performance of regression and artificial neural network schemes is evaluated for dynamic model reduction of power systems. The evaluation criterion is based on the goodness of fit in each reduced model with respect to the original model. Multiple linear regression, polynomial regression, and support vector are used as regression models while a Feedforward Artificial Neural Network with different activation functions is used for comparison with regression models. All simulations are based on a simplified Australian 14 Generator model. Datasets for training and test sets are obtained by measuring boundary bus properties and power flowing through tie lines. The simulation results show that the artificial neural network outperforms the regression models in making a reduced model of the power system, but only related to the system responses corresponding to the contingencies that were used for training. However, they perform poorly for unknown contingencies. Research work is being continued by the authors to create better models by combining classical models with machine learning techniques.