A class of randomized strategies has been proposed in [1-3] for selecting the desired model for a statistically steady experiment E whose probabilistic model (?, ?, P(?)) is unknown and belongs to some family {(?, ?, Pu(?)): u ? U} of permutation distributions. The main result of these works is the study of the family features and the fundamental features of permutation and quasi-permutation strategies for making decisions on the experiment E probabilistic model. The determination of the control system probabilistic model [4-6] is reduced to the solution of some optimization problem with regard to the information on a single trial of the experiment E. Being a continuation of [1-3], this work studies the problem of construction and features of a probability model family for the evolutionary experiment [7, 8] to reduce the extreme risk function or the probability of making a wrong decision on the choice of the model.
|
