Design of self-organizing fuzzy neural networks using evolutionary algorithms

In this thesis, an improved fast and accurate online self-organizing scheme for parsimonious fuzzy neural networks (FAOS-PFNN) has been proposed. The existing FAOS-PFNN has been proved that it can achieve fast learning speed and more compact network, as it only has growing procedure and do not have pruning. However, it has two weaknesses: (1) it needs sufficient training data to train the network to obtain a satisfactory modelling result; (2) the modelling will fail if a set of linearly-spacing training data is present to the data. The proposed improved FAO-PFNN successfully overcomes the two weaknesses by approximating local maxima of the modelling function before applying extended Kalman filter to update the parameter of the network. The intensive simulation results show that the proposed algorithm performs very well in function approximation.Bachelor of Engineerin