The thesis is devoted to investigating neurocontrol of nonlinear systems with uncertain and unknown dynamic models. The aim of my research in the neural network area is to search for fast learning algorithm with reduced computation burden. Novel theoretical synthesis and analysis of neurocontrol systems have been conducted and applied to control a robotic manipulator. Modified back propagation learning algorithm making use of the changeable shape of the nonlinear function of node is introduced. The resulting algorithm results in a better accuracy and faster convergence. Neural network based control scheme is used to control the motion of a manipulator. The neural network plays the role of an approximate inverse model of a robot and are then used in conjunction with a conventional proportional plus derivative (PD) controller. To demonstrate the feasibility of the proposed algorithm and neurocontrol scheme, intensive computer simulations were conducted. Different types of adaptive tracking problems and regulation problems are considered. The proposed scheme possesses robustness to systems model uncertainty and changing conditions of operations. Simulation results are quite promising. It is concluded that neural networks, by virtue of their natural ability to learn from data, are well suited for dynamic reconstruction,, bringing the world of nonlinear dynamics closer to practical use.

To demonstrate the practicability of the proposed scheme, experiments were conducted on an existing two-link manipulator and a single link manipulator. Results confirm the practicability of the proposed scheme.

The thesis concludes that the neurocontrol approach is capable of learning the tasks of reasonable complexity and it should be possible to train a system for a variety of operations using a neural network of practical size.