The Research And Application Of The Fuzzy Neural Network

Fuzzy-neural control is an active research area of intelligent control theory. There are still some theoretical and practical problems although fuzzy-neural nets have been applied in complicated system control and modeling. The objective of this thesis is to address some of these problems, such as the learning ability of fuzzy-neural network, the generation of fuzzy rule, the integration approaches of fuzzy logic and neural network, the application of fuzzy-neural nets to complicated system identification and control.The major contributions of the thesis include:1. A survey on the application of fuzzy-neural network to complicated system identification and control is presented. The problems in its theory and applications are analyzed.2. A study on fuzzy-neural network is conducted according to the problems aforementioned. A new construction method for fuzzy-neural system is proposed based on the analysis and comparison of existing methods, which is able to generate the fuzzy rules automatically.3. A new system identification method based on fuzzy-neural nets is proposed using the new construction method Experimental results show that the proposed method improves the identification accuracy, accelerates the convergence of the back-propagation learning, and enhances the stability of system.4. An adaptive fuzzy-neural network-based generalized predictive control algorithm is developed for nonlinear system. The algorithm can be used to the nonlinear system which is constrained by the range and velocity of control signal. The optimization of the algorithm is based on golden section approach which is non-derivative-based. The search space is dynamically set according to the constraint conditions of the system. The simulation testing shows that the optimization is accelerated while the stability and robustness of system are guaranteed.5. A new construction approach is proposed for fuzzy-neural hybrid system. The neural network-based adaptive control and fuzzy logic are integrated based on feedback learning algorithm. Experimental results show that it is able to deal with the nonlinearity and uncertainty of system. Overshooting is hardly observed in the experiments. Furthermore, it can tolerate the unmatched model to some extent.