Structural and parametric synthesis of population algorithms for global optimization
Anatoly Karpenko, Inna Kuzmina · Procedia Computer Science · 2021
From year to year, the computational complexity and dimension of applied global optimization problems is growing rapidly. This trend requires the development of new and increased efficiency of well-known global optimization algorithms. One of the most effective classes of such algorithms are intelligent population algorithms (P-algorithms). Due to the large and ever-increasing number of such algorithms, the development of algorithmic and software for their automated synthesis is relevant. We show that P-algorithms have a similar modular structure. This makes it possible to obtain a large number of variants of algorithms by combining the operators of various algorithms. P-algorithms also contain of a large number of free parameters in them. So, its efficiency can vary significantly depending on their values, while formal rules for choosing the values of these parameters are usually absent. These circumstances are the basis for posing the problem of structural-parametric meta-optimization of P-algorithms. We present the formulation of the problem of structural-parametric synthesis of P-algorithms. We offer a methodology for solving this problem, as well as present a prototype software system that implements this technique. In addition, we show in the work some results of computational experiments. This work was supported by the RFBR grant 18-07-00341.