Globally Stable Adaptive Neural Network Control And Its Application

During the past decades, interest in the lower triangular nonlinear systems has been even increasing, and many significant developments have been achieved. In practice,uncertainty is one of important factors which affect the control performance and the closed-loop stability of the whole systems. Universal function approximators, including neural networks or fuzzy logic systems, etc., have been proved to be a promising tool to model the unknown uncertainties for their good approximation ability over a compact domain. As the main design technique of the lower-triangular systems, the Backstepping technique plays an important role. Thus, many so-called approximation-based adaptive Backstepping approach, combining the universal function approximators with the adaptive Backstepping technique, has been found to be particularly useful for nonlinear systems with triangular structure.In fact, there indeed exist some real-world systems that can be described by the above system, such as induction motor, mobile robot, helicopters and missile mechanical.Therefore, while the neural networks nodes increase, the number of adaptive parameters will increase significantly. We cannot find appropriate method to determine the RBF NN approximation domain when using conventional ANNC to control the mobile robots since it relies on the uncertain position information. This main results and contribution of this dissertation are as follows.Firstly, in this paper we focus on the problem of adaptive neural control for strict-feedback nonlinear system. With the help of backstepping technique, we proposed a novel adaptive neural scheme and guaranteed that all the signals in the closedloop systems are globally uniformly bounded. Compared with the existing results on the control of nonlinear systems, the main advantage of this paper lies in that the proposed control scheme requires only two adaptive parameter need to be estimated online for an n-order nonlinear strict-feedback system regardless of the number of the neural network nodes. In this way, the computational burden is significantly alleviated, and thus the proposed control law may easily implement in practical applications. At last, a simulation example is provided to verify the control approach proposed in this paper.Secondly, in this section a new adaptive position tracking control strategy is proposed for a class of wheeled mobile robot systems where radial basis function(RBF) neural network(NN) is used to model the uncertainty. The so-called feedforward compensation scheme is developed where only the information of the reference position is employed as the NN input. The main advantage is that the global stability of closed-loop systems can be guaranteed and the NN approximation domain can be determined based on the reference signal a prior, which is different from the conventional adaptive neural network control(ANNC) schemes where only the semi-globally stable result can be obtained and no method is provided to determined NN approximation domain. Finally, a simulation is given to verify the effectiveness of the proposed control scheme.