Development and Rubustness Analysis of Fifth-Level Adaptive Critics for Control

点击翻译按钮获取中文摘要

9634127 Balakrishnan This study will build on previous supported research into brain-like intelligent control, using an advanced form of neural network design. This research has three major objectives: the first is to develop two level-5 adaptive critic designs. (Increasing level numbers indicate increasing level of capabilities to solve engineering problems and closer mimicking of perceived human intelligence. Currently, the highest level of critic design that has been well understood and used in a few applications is a level- 4 critic.) The first critic design based on approximate dynamic programming outputs the optimal return function and the derivatives of cost with respect to system states. The second critic design outputs the Hamiltonian and the derivatives of the Hamiltonian with respect to states and control. Advantages of these designs include retrieving Ricatti mattrix from the critic, and being able to check the Legendre-Clebsch condition, necessary for minimality. The second major objective is to investigate the robustness of the synthesized controller to: a) parametric variations, b) bound but unmodelled disturbances, and c) stochastic disturbances. The robustness analysis will be based on singular value based techniques, Liapunov concepts, and Montecarlo simulations. The third major objective is to use a set of two critics in conjunction to solve bench mark engineering problems involving inner loop and outer loop to account for slow dynamics and fast dynamics. The significance or major goal of this research is to combine human problem solving (brain like intelligence) philosophy embedded in a level-5 critic design and firm mathematical foundations offered by modern control and dynamic programming to solve difficult control problems in engineering. This approach not only finds optimal solutions but attempts to analyze the resulting solutions through robustness analysis. Such relations to establish relative stability of neurocontrolle rs are expected to help in the acceptance of neurocontrol for implementation in engineering. *** ??