Decomposition and coordination methods for constrained optimization

Decomposition and coordination methods based on the utilization of the Lagrangian generalized gradients in multilevel control structures have been developed successfully to handle the class of constrained optimization problems with nonseparable performance indices. Introducing proper intervention parameters, the overall problem is decomposed into a group of lower-dimensional constrained problems. A class of three-level control structures is thus proposed to coordinate the local solutions via two coordinating levels. While the first coordinating level is basically designed to adjust the state and control constraint violations, the second coordinating level is responsible mainly for manipulating the subsystems interactions in different ways. Salient features, advantages, and limitations of the developed control structures are pointed out.