Abstract:

To forecast failure rates better, a method for failure rate forecasting based on artificial neural networks is advanced. The basic ideas, forecasting models and steps of failure rate forecasting based on back propagation(BP) network and radial basis function(RBF) network are discussed respectively. The differences between neural network and other statistical forecasting methods, such as regression analysis, decomposition analysis, moving average, exponential smoothing, selfadaptive filtering and autoregressive moving average(ARMA) model, are compared respectively. The conclusion is drawn that neural network is the best statistical method of them according to the characteristics of failure rates.The failure rates of Boeing flights in a certain airline company are forecasted according to these two models.Results show that both of these models are suitable for failure rate forecasting,errors between predicted value and observed value are less than 20%,and the effect of failure rate forecasting based on RBF network is better than that based on BP network.By compared to errors of those statistical methods,errors of failure rate forecasting based on neural networks are much smaller.

Key words: neural , network,  , back , propagation,  , radial , basis , function , network,  , reliability,  , forecasting