Efficient parallel algorithm for optimal block decomposition of transfer function matrices for decentralized control of large-scale systems

The purpose of this paper is to present a parallel algorithm which can be used to determine how a large-scale system can be best decomposed in simpler subsystems to which decentralized control can be applied. The algorithm is based on an analysis in the frequency domain which employs the notion of block-diagonal dominance. The decentralization design problem is set as a combinatorial optimization problem and a "branch and bound" approach is applied to solve it. The proposed algorithm is conceived for parallel computational architectures. A discussion on the computational efficiency gained by parallelization is presented, and the implementation of the algorithm on a transputer network is briefly illustrated.<<ETX>>