Heuristic learning in recurrent neural fuzzy networks.

*(English)*Zbl 1110.68432Summary: A novel recurrent neural fuzzy network is proposed in this paper. The network model is composed by two structures: a fuzzy system and a neural network. The fuzzy system contains fuzzy neurons modeled with the aid of logic and and or operations processed via t-norms and s-norms. The neural network is composed by nonlinear elements placed in series with the previous logical elements. The network model implicitly encodes a set of if-then rules and its recurrent multilayered structure performs fuzzy inference. The topology induces a clear relationship between the network structure and an associated fuzzy rule-based system. In particular we explore this structure with an heuristic learning algorithm based on associative reinforcement learning and gradient search. These learning algorithms are associated to the fuzzy system and neural network, respectively. That is, output layer weights are adjusted via an error gradient method whereas a reward and punishment scheme updates the hidden layer weights. The recurrent fuzzy neural network is particularly suitable to model nonlinear dynamic systems and to learn sequences. Computational experiments with system identification problems show that the fuzzy neural models learned are simpler and that learning is faster than its counterparts.