Artificial Neural Network-Based Transient Stability Regions of Power Systems

Stability analysis of a power system is of critical importance for the maintenance of a reliable commercial power supply in an interconnected power system. In this dissertation a new method is presented for recognizing the stability regions of several model power systems using artificial neural networks (ANN). The stability analysis used in this dissertation is based on Lyapunov's second method. The region defined by the Lyapunov function was used as the desired output for the neural network. The ANN was trained to predict the stability regions for the following cases: (1) a power system with and without damping (2) a power system with and without a first-order velocity governor, (3) a power system with transfer conductances, (4) linearized models of power system based on Cartwright's method and Aizerman's method, (5) a class of nonlinear control system, and (6) the critical clearing angle for a simple two-machine system is also predicted using an ANN. The results showed that for most examples a single hidden layer with approximately twelve neurons was the optimum architecture. The use of variously sized neural network training sets was examined. Ultimate ANN convergence was affected by the initial weighting function chosen. It was found that the optimum number of neurons in the hidden layer was more depended upon the complexity of the decision region than on the number of input variables.