Transient stability assessment of power systems described with detailed models using neural networks

This paper investigates the use of multi-layered perceptron (MLP) neural network (NN) for assessing the transient stability of a power system considering the detailed models for the synchronous machines, and their automatic voltage regulators (AVRs). Two MLP NNs are employed here to estimate the critical clearing time (CCT) and a transient stability time margin (TM), as indicators for measuring power system transient stability for a particular contingency under different system operating conditions. The training of MLP NNs is accomplished using some carefully chosen system features as the inputs and the CCT and/or the TM as the desired targets. In this paper, the required training and/or testing patterns for the neural network are obtained by performing time-domain simulation (TDS) on the New England 10-machine 39-bus test system and the IEEE 16-machine 68-bus test system with fourth-order machines models and their AVRs using the software tool PSAT (Power System Analysis Toolbox), whereas the proposed neural network models are implemented in MATLAB. In addition, a neural network based sensitivity method and principal component analysis (PCA) are employed to reduce the dimension of the input data vectors. The simulation results obtained prove that the trained neural networks give satisfactory estimations for both CCT and TM.

► Implantation of the detailed models of two power systems in PSAT has been described.
► Many simulations showing the inaccuracy of the classical model have been illustrated.
► A neural network-based sensitivity method has been used to reduce the dimension of the inputs.
► Besides operating conditions parameters related to AVR and exciter have been used as NNs inputs.