On the importance of training methods and ensemble aggregation for runoff prediction by means of artificial neural networks
Adam P. Piotrowski , Jarosław J. Napiórkowski , Marzena Osuch , Maciej J. Napiórkowski
AbstractArtificial neural networks (ANNs) become widely used for runoff forecasting in numerous studies. Usually classical gradient-based methods are applied in ANN training and a single ANN model is used. To improve the modeling performance in some papers ensemble aggregation approaches, in others novel training methods are proposed. In this study the usefulness of both concepts is analysed. First, the applicability of large number of population-based metaheuristics to ANN training for runoff forecasting is tested on data collected from four catchments, namely Upper Annapolis (Nova Scotia, Canada), Biala Tarnowska (Poland), upper Allier (France) and Axe Creek (Victoria, Australia). Then the importance of the search for novel training method is compared with the importance of the use of a very simple ANN ensemble aggregation approach. It is shown that although some metaheuristics may slightly outperform the classical gradient-based Levenberg-Marquardt algorithm for specific catchment, none performs better for the majority of tested ones. One may also point out few metaheuristics that do not suit for ANN training at all. On the other hand, application of even the simplest ensemble aggregation approach clearly improves the results when the ensemble members are trained by any among suitable algorithms.
|Journal series||Hydrological Sciences Journal-Journal Des Sciences Hydrologiques, ISSN 0262-6667|
|Publication size in sheets||1.1|
|Keywords in English||catchment runoff forecasting, artificial neural networks, ensemble averaging, evolutionary algorithms, differential evolution, population size|
|Score|| = 30.0, 28-11-2017, ArticleFromJournal|
= 30.0, 28-11-2017, ArticleFromJournal
|Publication indicators||: 2016 = 2.222 (2) - 2016=2.372 (5)|
|Accepted||author version posted online: 23 Oct 2015|
* presented citation count is obtained through Internet information analysis and it is close to the number calculated by the Publish or Perish system.