A study of the bootstrap method for estimating the accuracy of artificial neural networks in predicting nuclear transient processes

The quantification of the uncertainty associated to the results provided by artificial neural networks is essential for their confident and reliable use in practice. This is particularly true for control and safety applications in critical technologies such as those of the nuclear industry. In this paper, the results of a study concerning the use of the bootstrap method for quantifying the uncertainties in the output of supervised neural networks are reported. A thorough parametric analysis is performed with reference to a literature problem. A case study is then provided, concerning the prediction of the feedwater flow rate in a Boiling Water Reactor.