The new ARSTI optimization method: Adaptive random search with translating intervals

A universal optimization method of adaptive random search with translating intervals for finding of the local extremum of a many-parameter objective function is developed. The aim of the study is more effective use of results from previous steps, as the accumulated ‘experience’ is taken into account during determination of the current working step direction. As a result of adaptive random search at initial directions with equal probabilities, a predominant motion is realized in these directions which ensure an extremum of the chosen optimization criterion. Convergence of the proposed ARSTI method is investigated. A comparative analysis with other methods of random search is made with respect to the number of objective function calculations needed for reaching a given solution accuracy (number of iterations). A program in FORTRAN 77 is given.