Flow prediction by three back propagation techniques using k-fold partitioning of neural network training data

Abstract

Flow forecasting performance by artificial neural networks (ANNs) is generally considered to be dependent on the data length. In this study k-fold partitioning, a statistical method, was employed in the ANN training stage. The method was found useful in the case of using the conventional feed-forward back propagation algorithm. It was shown that with a data period much shorter than the whole training duration similar flow prediction performance could be obtained. Prediction performance and convergence velocity comparison between three different back propagation algorithms, Levenberg–Marquardt, conjugate gradient and gradient descent was the next concern of the study and the Levenberg–Marquardt technique was found advantageous thanks to its shorter training duration and more satisfactory performance criteria.