Control of large decentralized dynamic systems

The nature of information flow in arriving at optimal control signals is examined for decentralized linear dynamic systems with local and supervisory controllers. It is illustrated for problems with quadratic cost subject to terminal (target) state constraint. It is assumed that local controllers know only their own dynamics and subsystem state vectors and that the supervisory control knows the target state vector and the dynamic and control coupling terms that exist between the subsystems. It is further assumed that the controllers can exchange intermediate results of computations in deriving control laws for the decentralized systems.