ADAPTIVE DYNAMIC EQUILIBRIUMS IN PROBLEMS of
STOCHASTIC
CONTROL UNDER UNCERTAINTY
V.V. Baranov
The problem of
stochastic control and decisions making under uncertainty concerning to state and
transient function, definite on outcomes, and ambiguously of given function of
utility is considered. The base
apriory of data is determined by set of objects
CBG = {X, a0 (X),Y, [YxÌY, xÎX], Z,[XzÌ X,zÎZ],G,qg(Z|Z´X´Y),wg(Z´X´Y), gÎG},
where
X is set of states, inaccessible to
supervision; a0(X)
-the apriory distribution on X; Y - set of allowable controls; YxÌY - restriction on admission of controls in dependences on state xÎX; Z - set of random
outcomes; XzÌ X - restriction on admission of state as alternatives of their
diagnostic in dependence on outcomes zÎZ; qg (Z |Z´X´Y) - transient function,
determining probability of transitions from Z´X´Y to Z, which, however, is not
given; G - set of hypotheseses about transient function; w g (Z´X´Y) - function of utility,
presenting preferences on controls y ÎYx pursuant to condition: y¢
y Û wg (z, x, y¢) > w g (z, x, y).
The solution of problem is based on principle of decomposition: at
uncertainty the problem is solved by redution to set interdependent of tasks,
inducing by corresponding to nature of uncertainty, and in joint their
solutions.
In particularly, in conditions of
base CBG the problem is reduced to
three tasks: diagnostics of state,
identification of transient function and selection of rule of control.
The mutual dependence of such tasks
induces the game content of problem with unopposite interests.
The problem is considered in some
assumptions and designs.
From
received here results the following final statements follow:
1.
Problem of stochastic control at uncertainty in conditions of base CBG
is solvable in principle and constructionally.
2. Principle of decomposition and
structural transformations determine the effective
method of formalization the problem, inducing the game statement of problem
with unopposite interests, solution of
which is described by equilibrium strategy.
3. In considered conditions the
equilibrium strategy asymptotical stationary, adaptive and is computed
consecutively on trajectory of outcomes. Adaptability provides the convergence
of process to stationary equilibriums, achievable at known true transient
function.
The developed approach and methods
permit to solution enough the wide class of pra-ctical problems of control in complex
systems at various variants of uncertainty, in particular, task of directed
development, control of socio - economic processes and many other ones.