| Description |
vii, 42 leaves : color illustrations ; 29 cm |
| Summary |
"In this thesis, the optimal control of a hypersonic vehicle in ascent through the atmosphere is developed using neural networks. Using numerical methods of optimal control and the flexibility and nonlinear mapping ability of neural networks, such a controller of a highly nonlinear system is feasible. The full nonlinear equations of motion are modeled and along with the control requirements are modified from an open final time time-invariant optimal control problem to a terminal 'time'-variant optimal control problem. A study of the physics of the problem using modified Two Point Boundary Value Problem solutions is undertaken to provide a reference optimal control solution. A modified extension to the Adaptive Critic Neural Network architecture is developed specifically for the structure of this problem. A series of controllers trained through this neural network structure are formed which control the vehicle along an optimal trajectory for a range of initial conditions. These neurocontrollers need no external training, perform with a degree of robustness to the initial conditions and are directly capable of being implemented in feedback control form"--Abstract, leaf iii. |
|
