The accuracy of a neural network (NN) depends on the quality and quantity of training data. For a given NN structure, the accuracy of the output depends on the number of training data and their distribution. If the NN structure is fixed and the number of training data is kept constant then the accuracy of the NN output, in general, will depend on the distribution of the training data. As a result having an appropriate distribution of the data is very important. Sobol's method can be used to generate a quasi-random sequence, which can provide good coverage 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) function approximation for a 4-machine 11 bus system under variations in load, fault location and network structure The results when compared with time domain simulation methods show less mean absolute errors than using a pseudo random choice of inputs.
展开▼