Efficient Computation of the Best Controls in Complex Systems Under Global Constraints

The problem of efficient calculation of optimal controls in large-scale systems is considered. The structure of interconnections is assumed to be known, whereas parameters of individual components remain unknown. Particular elements cannot be optimized independently of the remaining part of the system owing to some global constraints connected with limited resources. Using input-output observations all blocks are firstly identified by recursive instrumental variables method. Next, high-dimensional optimization problem is decomposed on series of levels. On each level one-dimensional problem is solved with the use of convex programming, taking into account the amount of available resources. Lower levels play supporting role for upper ones.