NSNFRM: construction of neuro TSK new fuzzy reasoning model using hybrid genetic-least squares algorithm

In this paper, a neuro TSK new fuzzy reasoning model (NSNFRM) is proposed. The proposed model uses the antecedent part of the neuro-new fuzzy reasoning model (NNFRM) proposed by the authors in Mazhar Tayel et al. [2000] and the consequent part of the TSK fuzzy model. The proposed model combines the advantage of the NNFRM and the advantage of the TSK fuzzy model. The advantages of NNFRM is first, using a neural network to represent the fuzzy model and second, using small number of rules and inputs. The advantage of TSK fuzzy model is getting lower mean square error. The parameter of the proposed model is tuned by a proposed hybrid genetic-recursive least squares (RLS) learning algorithm in which the antecedent parameters are tuned by the genetic algorithm and the consequent parameters are tuned by RLS algorithm. The performance of the proposed model is evaluated using a benchmark problem and compared with other modeling methods. It is shown that the proposed model outperforms other modeling methods including the standard TSK fuzzy model and the NNFRM model.