Search

The paper gives a survey of some hierarchical, multilevel - multilayer methods having been developed resp. advanced in the Dept. of Aut. Control of the Ilmenau Institute of Technology for on-line control of disturbed complex dynamic processes. In the first part, there will be a presentation of suitable extensions of multilevel methods permitting a static optimization of interconnected dynamic subsystems being effected directly at the process or by means of quickly adapted process models. The second part contains a decomposition of the global dynamic optimization problemt thus permitting a theoretically based synthesis of multilayer structures for an optimal control of stationary and transient processes in a single dynamic system. The combined multilevel - multilayer control structures presented in the third part allow an on-line control of the stationary and transient behaviour of interconnected dynamic subsystems and constitute a basis for automatic control of large-scale systems. By constructing special process outputs it has been possible to reduce the dimensions of the optimization and coordination problems

For the optimal control of a linear system with measurable or observable disturbances according to a quadratic criterion it is possible (with assumptions which are reasonable for practical applicability) to decompose the central feedforward-feedback solution into a two-layer control consisting of a superordinate on-line static feedforward optimization (depending on the disturbances) and a subordinate optimal dynamic transition into the actual static-optimum state (by means of state feedback). These results hitherto obtained are extended in this paper to nonlinear dynamic processes of the Hammerstein-type. With regard to the implementation on a multi-microcomputer-system, time-discrete algorithms are exclusively taken into consideration. After setting up the optimal central feedforward-feedback control for a linear system, a multilayer algorithm is developed for the on-line control of a nonlinear Hammerstein-process by heuristically applying the decomposition derived for linear systems (explained above). After that, an optimal time-discrete state observer for a Hammerstein-process is constructed. Finally some explanations are given concerning the implementation of the multilayer algorithm on a multicomputer system comprising problems of the hardware structure and those of the software design.