## Revista Learning and NonLinear Models

A revista ** Learning & Nonlinear Models (L&NLM;) ** é uma iniciativa pioneira da Sociedade Brasileira de Redes Neurais (SBRN), cujo o objetivo é divulgar a produção científica acadêmica e profissional na área de Inteligência Computacional (IC) e Sistemas Não-Lineares (SNL). A revista L&NLM; aceita artigos tanto de viés teórico, quanto aqueles orientados a aplicações nas mais diversas áreas de abrangência da IC/SNL, tais como redes artificiais, sistemas nebulosos, computação evolucionária, inteligência de enxame, aprendizado de máquinas, mineração de dados, predição de séries temporais, modelagem de sistemas caóticos, identificação de sistemas, robótica, biomédica, detecção e isolamento de falhas, processamento de sinais e imagens, controle de sistemas não-lineares, dentre outros. Artigos de revisão do estado da arte de qualquer um dos tópicos supracitados são também de interesse para a revista.

## Artigos Mais Baixados

G. Quadrelli, R. Tanscheit e M. M. Vellasco

**Abstract**
**Download**
The main goal of this paper is to propose procedures for modelling and
control of nonlinear systems by using neuro-fuzzy topologies. For the
modelling of a nonlinear system, its input space is initially divided
into a number of fuzzy operating regions, within which reduced order
models represent the system’s behaviour. The complete system modelling
– the global model – is obtained through the conjunction of the local
models by using a neuro-fuzzy network. A neuro-fuzzy adaptive network,
based on a hybrid learning algorithm (self-organised learning and
supervised learning) and called FALCON-H, is used in the control of a
nonlinear plant modelled as described above.

**Abstract**
**Download**
The novel approach to artificial neural networks based on the design of
task-specific networks and on biological models of a neuron with
multiple synapses developed by Baptista, Cabral and Soares (1998) was
extended to accommodate external perturbations. The learning algorithm
of this artificial neural network is an unsupervised training method
based on the processes of habituation, sensitization and classical
conditioning of human reflexes. In this paper, this new development is
applied to the control of the fluid temperature at any point in a
natural circulation loop. The learning and the action processes were
implemented in a computer program. The thermal-hydraulics processes
were also simulated. The natural circulation loop simulation model is
based on physical equations and on experimentally identified
parameters. The results show that besides the excellent learning
capability and generalization, the new improvements are suitable to
accommodate external perturbations so that the network is able to
maintain the controlled variable within allowable limits even in the
presence of strong perturbations.

**Abstract**
**Download**
Research on genetic algorithms (GAs) has shown that the initial
proposals are incapable of solving hard problems in a robust and
efficient way. Usually, for large-scale optimization problems, the
execution time of first generation GAs dramatically increases while
solution quality decreases. The focus of thistutorial is pointing out
the main design issues in tailoring GAs to large-scale optimization
problems. Important topics such as encoding schemes, selection
procedures, self-adaptive and knowledge-based operators are discussed.