Adaptive Backstepping Fuzzy/Neural Networks Control Of MIMO Nonlinear Systems

In this thesis, adaptive fuzzy/neural control schemes are presented for a class of uncertain MIMO nonlinear systems. We consider the follow broader class of multi-input/multi-output (MIMO) nonlinear systems.Where are the states of the system, and are the system inputs and outputs, respectively;are unknown nonlinear smooth functions.The main special features or this thesis are as follows:(i) The complex of the system: The MIMO system is composed of subsystems which are interconnected. Compared with the MIMO system consided in [39], the MIMO system in this thesis is more general in system state interconnections, which makes it difficult to control the systems.(ii) The better result of the control: By exploiting the special properties of the affine terms of the MIMO system, the developed scheme avoids the controller singularity problem completely without using projection algorithms. By employing NNs to approximate all the uncertain nonlinear functions in the controller design, the developed scheme achieves boundedness of all the signals in the closed-loop of the MIMO systems. The tracking error converges to zeroasymptotically. In most of the backstepping design methodology, all the solutions are globally uniformly bounded.(Hi) The flexibility of design: as show in this thesis, the control design is not unique. In actual applications, different types of control maybe used for the best result.The thesis is brganized as follows:Chapter 1 introduces the basic concepts and methods of nonlinear adaptive control technology, especially on adaptive backsepping design. The chapter ends with the depiction of the primary research objectives of the thesis.In chapter 2, In order to avoid the difficulties of poor convergent and instability in neural control design, we propose a variational learning-rate in the neural network learning algorithm, the stability and convergence with faster speed can be guaranteed.In chapter 3, we consider adaptive neural control of the broader class of multi-input/multi-output (MIMO) nonlinear systems, which is composed of subsystems which are interconnected. Based on the universal approximation capability of fuzzy/neural networks, we design a novel adaptive backstepping method to achieve not only boundedness of all the signals in the closed-loop of the MIMO systems but also the tracking error converges to zero asymptotically.The final chapter summarizes the main results and makes conclusions of the dissertation. Some future research work is also described.