Fast critical clearing time estimation of a large power system using neural networks and Sobol sequences

The accuracy of a neural network (NN) is dependent upon the quantity and quality of training data. Weighted and weightless neural networks were used and compared. For a given NN, the accuracy of the output depends on both the number of input training data and their distribution. Sobol´s method can be used to generate a quasi-random sequence, which can provide a good distribution of the input data over specified ranges. The application of Sobol sequences (Sob) was applied to the selection of the training patterns of the critical clearing time (CCT) of a 10-machine 39 bus New England system under variations in load level, fault location and network structure. A pseudo random choice of the same number of NN training patterns was compared with Sob. The results indicate that the CCT can be effectively estimated using these NNs when trained with Sob