Adaptive Control Of Uncertain Nonlinear Systems Via Nonlinear Parameterized Fuzzy Neural Approach

Research on control of nonlinear systems with model uncertainty has great theoretical value and practical significance, and has become one of the research hotspots in the field of nonlinear system control. This paper presents nonlinear parameterized fuzzy neural adaptive control strategies, which provide effective solutions for the control of uncertain nonlinear system.For two classes of uncertain nonlinear systems, adaptive control method based on nonlinear parameterized fuzzy systems and direct adaptive control method based on extreme learning machine neural networks are proposed in this paper.First, a direct adaptive control based on fuzzy systems with variable contraction-expansion factors is proposed. Fuzzy systems with variable contraction-expansion factors are composed of general fuzzy basis functions which can contract and expand. By adjustment of contraction-expansion factors, the fuzzy partition of input space of fuzzy control systems is realized. Therefore, the control accuracy can be improved without adding fuzzy rules.Second,In order to streamline the fuzzy rules further, an adaptive fuzzy controller based on self-organized ellipsoidal basis functions is proposed. The self-organized ellipsoidal basis function fuzzy system is used to approximate ideal control law, and robust compensation is introduced to ensure stability of the system. In the self-organized ellipsoidal basis function fuzzy system, fuzzy rules can be grown and pruned dynamically which can simplify the structure of fuzzy systems and reduce the computation load without decreasing approximation accuracy. Simulation results indicate that the controller can achieve effective performance with fewer fuzzy rules.In the following part, an adaptive control method based on self-organized ellipsoidal basis function fuzzy system and backstepping method is proposed for a class of strict-feedback nonlinear systems with unmatched and unstructured uncertainty. In the controller, backstepping procedure is employed to design controllers, and self-organized ellipsoidal basis function fuzzy systems are used to approximate unknown nonlinear dynamics. Simulation results and comparisons demonstrate the effectiveness and superiority of the control method.Finally, two adaptive controllers based on extreme learning machine neural networks are proposed. Extreme learning machine has advantages of simple structure and fewer artificial adjusted parameters, so the adaptive extreme learning machine controller is simple to design and needs less computation. In order to reduce computational complexity further, an adaptive extreme learning machine control method based on minimum learning parameter algorithm is presented, in which only one parameter needs to be tuned online. And simulation results indicate the effectiveness and superiority of the method.