Analysis of the efficiency of a multi-step adaptive optimization algorithm with prediction

<jats:p>&lt;p&gt;&lt;strong&gt;Context and relevance.&lt;/strong&gt; Global extremum search problems arise in the solution of a wide range of applied optimization problems, including the tuning of technical system parameters and the study of complex multi-extremal functions. For such problems, the application of classical deterministic methods is often difficult due to the presence of local extreme, the complex structure of the objective function, and the lack of complete information about its properties. In this regard, the development of software tools implementing adaptive and metaheuristic optimization algorithms is of great relevance. &lt;strong&gt;Objective.&lt;/strong&gt; The aim of this work is to develop and study software for implementing a multi-step adaptive optimization algorithm with forecasting, designed to find the global extremum of the objective function within a given set of feasible solutions. &lt;strong&gt;Hypothesis.&lt;/strong&gt; It is hypothesized that using a predicted solution position, a history of successful steps, adaptive step size adjustment, and the generation of new initial solutions via the L&amp;eacute;vy distribution will improve the stability of the search and reduce the probability of prematurely reaching a local extremum. &lt;strong&gt;Methods and materials.&lt;/strong&gt; This paper considers a multi-step adaptive algorithm based on sequential search in the neighborhood of the predicted solution, the use of memory of successful positions, and the generation of new initial solutions using the L&amp;eacute;vy distribution. The developed software allows users to specify the objective function, the feasible solution region, and the algorithm parameters; perform computational experiments; and visualize the search process and the results obtained. Benchmark test functions of varying complexity were used to verify the program&amp;rsquo;s performance. &lt;strong&gt;Results.&lt;/strong&gt; The computational experiments conducted showed that the implemented algorithm allows finding solutions close to the exact values of the global extremum for both simple and multi-extremal test functions. The results obtained confirm the functionality of the developed software package and the possibility of its application for &lt;strong&gt;analysis of the impact of parameter selection&lt;/strong&gt; on the quality of the search. &lt;strong&gt;Conclusions. &lt;/strong&gt;The developed software can be used for the numerical study of global optimization problems, the analysis of the behavior of a multi-step adaptive algorithm, and the selection of its parameters. The use of prediction, adaptive step size adjustment, and the generation of new initial points makes it possible to improve the stability of the search and reduce the probability of premature convergence to a local extremum.&lt;/p&gt;</jats:p>