Research On Robust Adaptive Neural-net Based Control And Construction Of Distributed 3D Flight Simulation Platform

There are two main contents researched in this dissertation, which are robust self-adaptive neural net-based control and construction of distributed 3D flight simulation platform. Unlike traditional differential geometrical method, the robust adaptive neural net-based control emphasizes on the theory and application of intelligence control in nonlinear system that is weighed by robust performance index. The problem of robust reliability and stability in a kind of weak nonlinear system is first proposed in the dissertation, and then it deeply analyses and discusses the problem of how to design track controller in strong nonlinear system with uncertainties and external disturbance on the premise of guaranteeing the whole system's robust performance index. The distributed 3D flight simulation platform is mainly used to validate real-time program and evaluate the efficiency of advanced flight control laws, therefore, some theories and techniques related to the platform are also introduced.This dissertation presented two robust control methods for nonlinear system based on the technique of Linear Differential Inclusion (LDI). Above all, the design problem of robust reliable controller is discussed, which is in weak nonlinear time-lagged system. Because the extension of dynamic change in weak nonlinear system is not large, the robust reliable controller designed by LDI can make the whole controlled system stable when time lag and faults exit in the system, at the same time, satisfying robust performance index of the system. Next, considered that the approximation error produced by LDI, the unmodeling error produced by system, the parameter uncertainties and the external disturbances can not be ignored, a dynamic neural network controller is designed to compensate their effect on line. Adjusted by the state output error between the ideal model and the controlled system, the cooperation of on-line network compensator and linear H_{∞}controller of ideal model makes the whole close-loop system guarantee robust stability and track the specified signal well.This dissertation presented two new methods of robust adaptive track control for a class of MIMO strong nonlinear system with external disturbance. One method makes use of Taylor approximation principle to linearize the MIMO strong nonlinear system at the ideal equilibrium point, meanwhile external disturbance is considered, and then designs two on-line neural network controller respectively, which can dynamically compensate the high order items of Taylor series and the control signals at ideal equilibrium point under the drive of state error between linear and nonlinear system. A linear feedback controller obtained by Pole Assignment and two on-line neural network act onthe practical MIMO high nonlinear system together, guaranteeing the whole system robust stable and tracking the specified signal; the other method designs three on-line neural networks for this class of system. They respectively real-time counteract the nonlinear part, the nonlinear items coupled with control signals and the external disturbance, so that the outputs of controlled system can completely track the inputs of specified reference signals. These two methods are both validated by simulating a nonlinear aircraft model with six free degrees and twelve orders. The result shows that this kind of adaptive control method has robust performance to a great extent because of the special dynamic compensation for external disturbance in these two systems.Lastly, this dissertation introduces the main development techniques used for exploring distributed 3D flight platform which can validate advanced flight control laws in theory and method. Based on Windows 2000/XP OS (operation system), it synthetically applies real-time simulation arithmetic of control system, OOP (object oriented program) development technique, TCP/IP protocol-based net- communication technique and API development technique of 3D virtual reality OpenGL to exploring the distributed simulation platform with friendly a...