Flashover forecasting on high-voltage insulators with a backpropagation neural net

The flashover forecast problem on high-voltage insulators is addressed with a backpropagation neural network paradigm. The importance of analysis of artificial neural networks (ANNs) dedicated to forecast problems is emphasized. This analysis includes adequate training-data management and optimum topology design. The direct (or automatic) design approach is compared with the design-with-analysis approach, and the results are presented as seven guidelines for new designs of ANNs for prediction.