In the paper a state filtration in a decentralized discrete time Linear Quadratic Gaussian problem formulated for a multisensor system is considered. Local optimal control laws depend on global state estimates and are calculated by each node. In a classical centralized information pattern the global state estimators use measurements data from all nodes. In a decentralized system the global state estimates are computed at each node using local state estimates based on local measurements and values of previous controls, from other nodes. In the paper, contrary to this, the controls are not transmitted between nodes. It leads to nonconventional filtration because the controls from other nodes are treated as random variables for each node. The cost for the additional reduced transmission is an increased filter computation at each node.
In the paper, state filtration in a LQG problem formulated for a multisensor system is considered. Control is determined by a central node as a linear form of a state estimate. It is assumed that control values are not available to local nodes. Because of the drawbacks of centralized filtration an optimal fusion of decentralized local Kalman filters is proposed. When control values are not available to local nodes, then control should be treated as a random variable in the synthesis of local state estimates. This leads to a non-classical estimation. It is shown that the proposed filter is equivalent to the centralized one.
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.