Optimal design of artificial neural networks by a multi-objective strategy: groundwater level predictions
- 1 June 2006
- journal article
- research article
- Published by Taylor & Francis in Hydrological Sciences Journal
- Vol. 51 (3) , 502-523
- https://doi.org/10.1623/hysj.51.3.502
Abstract
Currently, environmental modelling is frequently conducted with the aid of artificial neural networks (ANNs) in an effort to achieve greater accuracy in simulation and forecasting beyond that typically obtained when using solely linear models. For the design of an ANN, modellers must contend with two key issues: (a) the selection of model input and (b) the determination of the number of hidden neurons. A novel approach is introduced to address the optimal design of ANNs based on a multi-objective strategy that enables the user to find a set of feasible ANNs, determined as optimal trade-off solutions between model simplicity and accuracy. This is achieved in a multi-objective fashion by simultaneously minimizing three different cost functions: the model input dimension, the hidden neuron number and the generalization error computed on a validation set of data. The multi-objective approach is based on the Pareto dominance criterion and an evolutionary strategy has been employed to solve the combina... Résumé Aujourd'hui, la modélisation environnementale est fréquemment abordée au moyen de réseaux de neurones artificiels (RNA), dans le but d'améliorer la précision des simulations et des prévisions, au delà de celle qui est généralement obtenue avec des modèles purement linéaires. Pour construire un RNA, le modélisateur doit aborder deux aspects clefs: (a) la sélection des données d'entrée et (b) la détermination du nombre de neurones cachés. Une nouvelle approche est introduite afin d'étudier la construction optimale de RNAs grâce à une stratégie multi-objectifs qui permet à l'utilisateur de trouver un ensemble de RNAs possibles, correspondant à des solutions optimales de compromis entre simplicité et précision du modèle. Cela est réalisé dans un cadre multi-objectifs via la minimisation simultanée de trois fonctions de coût différentes: la dimension des entrées du modèle, le nombre de neurones cachés et la généralisation de l'erreur calculée avec un jeu de données de validation. L'approche multi-object...Keywords
This publication has 27 references indexed in Scilit:
- Managing uncertainty in hydrological models using complementary modelsHydrological Sciences Journal, 2003
- Artificial neural network approach to flood forecasting in the River ArnoHydrological Sciences Journal, 2003
- Modelling and forecasting of hydrological variables using artificial neural networks: the Kafue River sub-basinHydrological Sciences Journal, 2003
- A fast and elitist multiobjective genetic algorithm: NSGA-IIIEEE Transactions on Evolutionary Computation, 2002
- Input–output dynamic neural networks simulating inflow–outflow phenomena in an urban hydrological basinJournal of Hydroinformatics, 2000
- Artificial Neural Networks in Hydrology. II: Hydrologic ApplicationsJournal of Hydrologic Engineering, 2000
- Using pruning algorithms and genetic algorithms to optimise network architectures and forecasting inputs in a neural network rainfall-runoff modelJournal of Hydroinformatics, 1999
- An artificial neural network approach to rainfall-runoff modellingHydrological Sciences Journal, 1998
- Multiobjective optimization and multiple constraint handling with evolutionary algorithms. I. A unified formulationIEEE Transactions on Systems, Man, and Cybernetics - Part A: Systems and Humans, 1998
- Bootstrap Methods: Another Look at the JackknifeThe Annals of Statistics, 1979