Large scale system decentralized controller design based on locally imposed decomposition constraints

This paper proposes a new method for the design of decentralized controllers for large scale systems. The controllers are designed in two steps. In the first step, a powerful assignment algorithm is used to pair the system's inputs and outputs and to identity clusters of strongly interacting subsystems. Then, a decomposition condition is derived which allows for the controller of each of these subsystems to be designed separately. Thus the original problem of designing the decentralized controller becomes that of solving several smaller local problems subject to the satisfaction of a local condition. Consequently each sub-controller can be designed and tuned separately, which is a useful feature in process control applications where decentralized structure is important. It is shown that both the local controller design and the satisfaction of the local decomposition condition can be combined together into a single convex optimization problem which can be solved by linear matrix inequalities.