A theoretical study on the synthesis of hierarchical neural controllers based on power systems is presented in the paper. This theoretical study concludes that the proposed hierarchical neural controller construction is not only useful for power system applications, but should be suitable for other applications. It first shows that time-optimal (or other "optimal") neural controllers can be synthesized to approximately identify the switching manifold for control. It then shows that the hierarchical neural controller call deal with system uncertainties, and should perform reasonably well in theory. Further, adaptive hierarchical neural controllers are developed to deal with time varying characteristic of the systems, and it is shown that they are able to perform robustly.