Institute of Cybernetics at Tallinn University of Technology            RESEARCH OVERVIEW 1998
next top of this page previous

logo DEPARTMENT OF MECHANICS AND APPLIED MATHEMATICS


STOCHASTIC PROGRAMMING

HEAD OF THE PROJECT Riho LEPP, Ph.D.

DESCRIPTION

Two types of stochastic programming (SP) models are widely known: two-stage SP-s (or stochastic programs with recourse) and chance-constrained (or probabilistic) SP-s. Last ones include the probability function v (x, fii) of the form

v (x, fii) = P { f (x, ksii) < fii} (1)

where ksii is a random parameter and x is the control parameter. The problem is to maximize v (x, fii) over x for a fixed level fii.

The inverse to the problem (1), the quantile minimization problem is defined in such a way that the probability level alfa, 0 < alfa < 1, is fixed earlier, and the purpose now is to minimize the quantile fii:

windex:alfa(x) = minindex:fii { v (x, fii) greater or equal alfa } ,(2)

and then to minimize windex:alfa(x) over x.

Both problems (1), (2) are mathematically quite uncomfortable to handle: v (x, fii) and windex:alfa(x) are never convex, only under very hard restrictions to distribution of ksii and to function f (x,ksii) they are differentiable etc. At the same time functions v (x, fii) and windex:alfa(x) are very important in applications. Consider as an example the problem of the correction of a satellite orbit.

The mathematical model of the correction is given by the following recursion relation: d (x,ksii) = ksii0 + x (1 + ksii1) where d is the drift speed after the correction, ksii0 is the current drift speed at the apogee, x is the drift speed impulse and ksii1 is the random coefficient characterizing the error of the correction impulse implementation. The variable ksii0 is the rest of the drift speed after the correction process at the previous stage. It is assumed that ksii0 and ksii1 are independent Gaussian random variables. The quantile function windex:alfa(x) is now defined as follows:

windex:alfa(x) = minindex: fii { fii : P (ksii | | d (x,ksii) | greater or equal alfa }(3)

and the quantile minimization problem is for this model the following: minimize over certain class of strategies x = x(ksii) the quantile function windex:alfa(x) where alfa is near to 1, e.g. alfa = 0,9999.

In general, there are two ways to find an optimal strategy: mean square and minimax ones. If the fixed probability level alfa is around 0,9 then differences between the two strategies are quite small, but in the case of alfagreater or equal0,999 the situation changes drastically, and we must use the more complicated, minimax strategy in order to find optimal quantile for the correction of the satellite orbit, see fii upper index E and fii upper index PII of mean square and minimax straregies in the figure for correction of the satellite orbit (3), respectively.



next top of this page previous 10/04/1998 webmaster

© Institute of Cybernetics, Akadeemia 21, Tallinn 12618, Estonia     Phone: (+372) 620 4150   Fax: (+372) 620 4151