Dynamic Programming and Stochastic Control
Dynamic Programming and Stochastic Control
| Hi-index | 22.14 |
The objective of this paper is to present a ''piece-by-piece'' LQG design for large-scale systems. The proposed design procedure is developed for a hierarchical (block-triangular) representation of the system, which is obtained via a graph-theoretic decomposition algorithm. The estimator is built as a union of low-order optimal estimators attached to each individual subsystem sequentially going from the top to the bottom of the hierarchy. As the subsystem state estimates become available, optimal controllers can be designed for each subsystem separately resulting in an overall closed-loop system which is stable and suboptimal. This design process offers a considerable reduction of both off-line and on-line computations, which is especially effective in large sparse systems.