Abordagem neurofuzzy para modelagem de sistemas dinamicos não lineares / Neurofuzzy approach for nonlinear dynamical systems modeling

AUTOR(ES)
DATA DE PUBLICAÇÃO

2008

RESUMO

This work suggests a systematic procedure to develop models of complex nonlinear dynamical systems using neural fuzzy networks. The neural fuzzy networks are able to extract knowledge from input/output data and to encode it explicitly in the form of if-then rules. Therefore, linguistic models are obtained in a form suitable for human understanding. Two new classes of fuzzy neurons are introduced to generalize AND and OR logic neurons. These generalized login neurons, called unineurons and nullneurons, provide a mechanism to implement synaptic plasticity and an important characteristic of biological neurons, the neuronal plasticity. Unineurons and nullneurons achieve synaptic and neuronal plasticity modifying their internal parameters in response to external changes. Thus, unineurons and nullneurons may individually vary from a AND neuron to a OR neuron (and vice-versa), depending upon the necessity of the modeling task. Neural fuzzy networks constructed with unineurons and nullneurons are more general than similar fuzzy neural approaches suggested in literature. Training algorithms for neural fuzzy networks with unineurons and nullneurons are also studied. In particular, a new training algorithm based on the participatory learning is introduced to develop fuzzy models of dynamical systems. In the participatory learning algorithm, a new information brought to the network through an input/output data is first compared with the knowledge that the network already has about the model. The new information influences the update of the knowledge only if it does not conflict with the current knowledge. As a result, neural fuzzy networks trained with participatory learning show greater robustness to training data with anomalous values than their counterparts. The neural fuzzy network and training algorithms suggested herein are used to develop time series forecast models and thermal models of power transformers. Experimental results show that the models proposed here are more robust and perform best in terms of accuracy and computational costs when compared against alternative approaches suggested in the literature

ASSUNTO(S)

neural networks (computer science) neurofuzzy systems sistemas nebulosos redes neurais (computação) modelos matematicos inteligencia artificial artificial intelligence mathematical models

ACESSO AO ARTIGO

Documentos Relacionados