| Summary: | There are a large number of examples of using the population-based algorithms (P-algorithms) to have a successful solution for complex practical tasks of global optimization, for instance, problems of computer-aided design, synthesis of complex chemical compounds, optimal control of dynamic systems, etc. P-algorithms are also successfully used in multi-criteria optimization, when a preliminary construction of some Pareto set (front) approximation is required. P-algorithms are numerous and very diverse – over 100 such algorithms are known, and new algorithms continue to appear. In this connection, the problem of systematizing expressive means of P-algorithms is of relevance. We consider one of the components of this problem that is the problem of classification of the search operators of P-algorithms.The paper formulates a global optimization problem and a general scheme of the P-algorithms to solve it. This multilevel classification of the main search operators of P-algorithms at the highest level of the hierarchy identifies the following operators: initialization of the population and the end of the search; coding of individuals; randomization; selection; crossing; population management; local search.These operators at the next hierarchical level are divided into deterministic and stochastic ones. Further, we distinguish static, dynamic program and dynamic adaptive operators. The following classification levels are "operator dependent", that is, generally speaking, different for each operator. We reveal the essence of these operators and give the use cases in various P-algorithms.Although the paper uses the names of operators such as selection, crossing, our orientation is not only to evolutionary algorithms. A description scheme of operators presented in this paper can be used to determine any population-based algorithms.The work development expects extending a set of operators presented, and, above all, using this set and a set of basic essences of population algorithms, the formalization of which was earlier proposed by the author, systematizing the most known algorithms of this class.
|
|---|