A methodology based on neural networks for the determination of the critical clearing time of power systems transient stability

In this paper, the authors propose a methodology based on artificial neural networks (ANN) trained conveniently to compute the critical clearing time (CCT) of power system transient stability. The selected ANN is a feedforward multilayer perceptron (MLP). The critical clearing time is the only variable considered in the output. The input variables of the neural network were selected after careful assessment and they are composed of the electrical network parameters and topology represented by elements of the admittance matrix, the amplitude and bandwidth corresponding to the first large oscillation of the swing curve power vs. angle of the advanced generator. Training of the ANN has been enhanced in terms of speed and precision with the adoption of the second order Levenberg-Marquardt optimization method. After training, the ANN is used in such manner that the input data is obtained with only one time-domain study of any fault and clearing time, then the neural network will estimate the CCT. The proposed methodology is valid not only for the first-oscillation transient stability analysis but also to calculate the CCT of power systems that may experience several over oscillations. The New England 10-generator system has been used to test the proposed ANN-based methodology. Results of the tests indicate that the maximum testing error in the calculation of the critical clearing time is below 5%.