@ARTICLE{26583204_97091013_2013, author = {Maxim Hivintsev and Andranik Akopov}, keywords = {, simulation modeling, distributed calculations, genetic algorithmsmulti-objective optimization}, title = {Distributed evolutionary network for the solution of multi-objective optimizing problems in simulation systems}, journal = {}, year = {2013}, number = {3(25)}, pages = {34-40}, url = {https://bijournal.hse.ru/en/2013--3(25)/97091013.html}, publisher = {}, abstract = {Maxim Hivintsev - Post-Graduate Student, Department of Business Analytics, Faculty of Business Informatics, National Research University Higher School of Economics.Address: 20, Myasnitskaya str., Moscow, 101000, Russian Federation.E-mail: mkhivintsev@hse.ruAndranik Akopov - Professor, Department of Business Analytics, Faculty of Business Informatics, National Research University Higher School of Economics.Address: 20, Myasnitskaya str., Moscow, 101000, Russian Federation.E-mail: aakopov@hse.ruOne of the problems related with optimization tasks is that simultaneous optimization on a set of criteria (decision variables) requires considerable time and computing resources. Thus, as objective functions of many (more than 10) criteria generate multidimensional space of possible solutions, computing complexity of an optimizing task, as a rule, becomes a barrier to its solution within acceptable time. Analytical description and precise solution of such large-scale tasks in practice, as a rule, are not possible.The purpose of the paper is development of the distributed evolutionary network intended for the solution of multi-criteria large-scale optimizing tasks. A new approach to developing of complex software solution regarding the integration of the distributed evolutionary network with simulation software AnyLogic is presented.In this paper a novel approach to the solution of multi-criteria large-scale optimization tasks, in particular, in simulation systems (for example, AnyLogic) using distributed calculations is presented. The new concept of developing distributed evolutionary network is based on splitting of space of decision variables into clusters. Each computing element of the network is assigned to a particular cluster, then intermediate results are obtained within this cluster using interacting genetic algorithms. The approach is based on a distributed evolutionary network in which the effect from parallelization of computing processes is higher, than in classical "island model". It is obtained by the ability of the algorithm to split the whole decision space on clusters, each of which is transferred to own computing process, with subsequent executing of genetic algorithm at local level. In this case the genetic algorithm operates with objective functions that have smaller number of decisions variables. The smaller size of population is required for obtaining acceptable decisions. As a result, the number of required recalculation of fitness functions decrease within the same time, and convergence of the algorithm to the final solution is significantly reduced.A platform-independent system based on proposed approach using Java programming language is developed. It allows to integrate of the distributed evolutionary network and simulation models developed in AnyLogic.}, annote = {Maxim Hivintsev - Post-Graduate Student, Department of Business Analytics, Faculty of Business Informatics, National Research University Higher School of Economics.Address: 20, Myasnitskaya str., Moscow, 101000, Russian Federation.E-mail: mkhivintsev@hse.ruAndranik Akopov - Professor, Department of Business Analytics, Faculty of Business Informatics, National Research University Higher School of Economics.Address: 20, Myasnitskaya str., Moscow, 101000, Russian Federation.E-mail: aakopov@hse.ruOne of the problems related with optimization tasks is that simultaneous optimization on a set of criteria (decision variables) requires considerable time and computing resources. Thus, as objective functions of many (more than 10) criteria generate multidimensional space of possible solutions, computing complexity of an optimizing task, as a rule, becomes a barrier to its solution within acceptable time. Analytical description and precise solution of such large-scale tasks in practice, as a rule, are not possible.The purpose of the paper is development of the distributed evolutionary network intended for the solution of multi-criteria large-scale optimizing tasks. A new approach to developing of complex software solution regarding the integration of the distributed evolutionary network with simulation software AnyLogic is presented.In this paper a novel approach to the solution of multi-criteria large-scale optimization tasks, in particular, in simulation systems (for example, AnyLogic) using distributed calculations is presented. The new concept of developing distributed evolutionary network is based on splitting of space of decision variables into clusters. Each computing element of the network is assigned to a particular cluster, then intermediate results are obtained within this cluster using interacting genetic algorithms. The approach is based on a distributed evolutionary network in which the effect from parallelization of computing processes is higher, than in classical "island model". It is obtained by the ability of the algorithm to split the whole decision space on clusters, each of which is transferred to own computing process, with subsequent executing of genetic algorithm at local level. In this case the genetic algorithm operates with objective functions that have smaller number of decisions variables. The smaller size of population is required for obtaining acceptable decisions. As a result, the number of required recalculation of fitness functions decrease within the same time, and convergence of the algorithm to the final solution is significantly reduced.A platform-independent system based on proposed approach using Java programming language is developed. It allows to integrate of the distributed evolutionary network and simulation models developed in AnyLogic.} }