Multilayer dynamic fuzzy neural network design for control and system identification applications

In this thesis, three different new neuro-fuzzy neural network models have been proposed for system identification and control problems. The structure of introduced models comes from the idea of adaptive neuro-fuzzy inference system (ANFIS) which is used for obtaining fuzzy rule base from the input-output data of an unknown function. The first model is type-2 fuzzy wavelet neural network (T2FWNN). In the T2FWNN structure, membership functions of fuzzy rules are represented with type-2 wavelet functions which have time and frequency localization properties, unlike the conventional fuzzy sets. In the consequent part of the rules, the weighted summation of wavelet functions is used. The second model is dynamic adaptive neuro-fuzzy inference system (DANFIS). DANFIS is a neural network representation of Takagi-Sugeno affine fuzzy model that incorporates local linear dynamics with bias terms into consequent parts of each fuzzy rule. Thus, the internal behavior of a nonlinear dynamical system will be able to be modeled with DANFIS. In DANFIS, Gaussian functions are used as membership functions and consequent part of the rules are represented with linear differential equation of the inputs. The suitable parameters of DANFIS are determined by gradient based training algorithm and adjoint sensitivity method. In the third model, wavelet functions are used as membership function different than DANFIS model and new network is called as dynamic fuzzy wavelet neural network (DFWNN). Simulation results are given to demonstrate the effectiveness of the proposed T2FWNN, DANFIS, DFWNN models for system identifications and control problems.