Services on Demand
Journal
Article
Indicators
- Cited by SciELO
- Access statistics
Related links
- Cited by Google
- Similars in SciELO
- Similars in Google
Share
Tecnura
Print version ISSN 0123-921X
Abstract
VILLARREAL LOPEZ, Edwin and ARANGO, Daniel Alejandro. Training strategies for fuzzy number neural ne. Tecnura [online]. 2014, vol.18, n.40, pp.36-47. ISSN 0123-921X.
Abstract The purpose of this article is to present general training strategies for training fuzzy number neural networks used in the learning of systems from linguistic data. We shortly analyze the main trends in the training of this kind of systems and from that point, we propose new strategies. The first of them is based on the backpropagation of the mean square error in all the α-cuts for crisp weights. The second strategy uses a real codification genetic algorithm for crisp weights networks. The third is based in the backpropagation of the mean error value and the ambiguity of all the α-cuts for fuzzy weights, and the last one uses the backpropagation of a fuzzy error measure for a fuzzy weighted network. An experimental stage is performed implementing the developed algorithms together with one of the most representative reported, allowing identifying the best suited data set for each of them. Finally the strategies are applied for an environmental impact assessment system for landfills.
Keywords : Backpropagation; Fuzzy Neural Networks; Genetic Algorithms.