Тип публикации: доклад, тезисы доклада, статья из сборника материалов конференций
Конференция: Advances of science; Karlovy Vary - Moscow; Karlovy Vary - Moscow
Год издания: 2018
Ключевые слова: global optimization, continuous variables, discrete variables, selective averaging of required variables, multiextreme function, inequality restrictions, continuous auxiliary variables
Аннотация: The idea of constructing of a basic algorithm of global optimization on a set of continuous and discrete variables with the ordered possible values at the presence of inequality restrictions is started. The algorithm is based on separation in time of trial and working steps, selective averaging of required variables by results of tПоказать полностьюrial movements, adaptive reorganization of the sizes of set of possible trial movements and the accounting of inequality restrictions. The main existing scheme of continuous variables search (with selective averaging) is transferred on mixed variables. For each discrete variable is assigned the continuous auxiliary dimensionless variable. It contains numbers of possible values of a discrete variable and the identical individual intervals, adjoining each other, covering these numbers. The numbers are located in the center of intervals. The auxiliary variable on each working step has its own fixed interval of possible values. When generating of each sampling point for the corresponding discrete variable on the continuous interval the sensor of uniform distributed continuous random value calculates a pseudo-random number, which gets on one of subintervals in the center of which there is number of the discrete value. Next, discrete value with this number participates in calculation of minimized function and of inequality restriction functions. Calculation for each discrete variable of the new center of search movements and the sizes of a set of search in the beginning is conducted on a continuous auxiliary variable and then transition to discrete possible values is carried out. Dimension of the problem of optimization at such approach does not change. Complexity of calculations in comparison with existence of only continuous variables increases nonessential. The example shows possibility of reception of a near-to-one the probability estimation of hitting in small neighborhood of true .
Журнал: Advances of science
Номера страниц: 80-92
Издатель: Международный центр научно-исследовательских проектов