Please use this identifier to cite or link to this item:
|Scopus||Web of Science®||Altmetric|
|Title:||Input determination for neural network models in water resources applications. Part 1 - background and methodology|
|Citation:||Journal of Hydrology, 2005; 301(1-4):75-92|
|Publisher:||Elsevier Science BV|
|Gavin J. Bowden, Graeme C. Dandy and Holger R. Maier|
|Abstract:||The use of artificial neural network (ANN) models in water resources applications has grown considerably over the last decade. However, an important step in the ANN modelling methodology that has received little attention is the selection of appropriate model inputs. This article is the first in a two-part series published in this issue and addresses the lack of a suitable input determination methodology for ANN models in water resources applications. The current state of input determination is reviewed and two input determination methodologies are presented. The first method is a model-free approach, which utilises a measure of the mutual information criterion to characterise the dependence between a potential model input and the output variable. To facilitate the calculation of dependence in the case of multiple inputs, a partial measure of the mutual information criterion is used. In the second method, a self-organizing map (SOM) is used to reduce the dimensionality of the input space and obtain independent inputs. To determine which inputs have a significant relationship with the output (dependent) variable, a hybrid genetic algorithm and general regression neural network (GAGRNN) is used. Both input determination techniques are tested on a number of synthetic data sets, where the dependence attributes were known a priori. In the second paper of the series, the input determination methodology is applied to a real-world case study in order to determine suitable model inputs for forecasting salinity in the River Murray, South Australia, 14 days in advance. © 2004 Elsevier B.V. All rights reserved.|
|Keywords:||Artificial neural networks|
General regression neural network
|Appears in Collections:||Aurora harvest 6|
Civil and Environmental Engineering publications
Environment Institute publications
Files in This Item:
There are no files associated with this item.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.