Bayesian estimation for decision-directed stochastic control

Bayesian estimation for decision-directed stochastic control

Blom, H.A.P.

National Aerospace Laboratory NLR

1990-03-15

Stochastic processes with a decision-directed control are considered as controlled Markov processes, the state space of which is hybrid; i.e. a product of a discrete set and a Euclidean space. This approach yields a mathematical model for many problems of decision-directed stochastic control. In general, the observations made from the "past" and "present" Markov state do not lead to a perfect knowledge of the "present" discrete-valued state component. In such situations, the optimal control may be obtained by applying two successive steps: - Bayesian estimation (evaluation of the conditional distribution) of the Markov process, - Optimal control of the conditional distribution on the basis of perfect knowledge of its evolution. Unfortunately, the evaluation of each of these steps implies significant difficulties in case the Markov state is hybrid. The thesis is directed to the modelling of hybrid state Markov processes and to solving problems that are associated with the Bayesian estimation of these processes.

martingales
stochastic processes
Markov processes
algorithms
smoothing
mathematical models
optimal control
decision theory
state estimation
nonlinear filters
tracking filters

http://resolver.tudelft.nl/uuid:d7711415-bf6c-4f87-87d1-4f1b537ca28a

Nationaal Lucht- en Ruimtevaartlaboratorium

Campus only

NLR Technical Publication TP 90039 U

Aerospace Engineering Reports

report

(c) 1990 National Aerospace Laboratory NLR