Sequential training of bootstrap aggregated neural networks for nonlinear systems modelling

Abstract

A sequential training method for developing bootstrap aggregated neural network models is proposed in this paper. In this method, individual networks within a bootstrap aggregated neural network model are trained sequentially. The first network is trained to minimise its prediction error on the training data. In the training of subsequent networks, the training objective is not only to minimise the individual networks' prediction errors but also to minimise the correlation among the individual networks. Training data sets for the individual networks are different and are generated through bootstrap re-sampling of the original training data set. Training is terminated when the aggregated network prediction performance cannot be further improved. An application example demonstrates the superior performance of this neural network training strategy.