Другие журналы
|
scientific edition of Bauman MSTUSCIENCE & EDUCATIONBauman Moscow State Technical University. El № FS 77 - 48211. ISSN 1994-0408
Multi-memetic Mind Evolutionary Computation Algorithm for Loosely Coupled Systems of Desktop Computers
# 10, October 2015 DOI: 10.7463/1015.0814435
Article file:
SE-BMSTU...o452.pdf
(812.58Kb)
Solving complex optimization problems in the field of engineering design is quite often demand large computational expenses. Nowadays, distributed parallel computations are one of the most promising approaches to raise the efficiency of methods for dealing with this type of problems. At the same time, it would not be enough just to increase the amount of computational resources for obtaining high-quality solutions to optimization problems. A development of specialized algorithms, which would be based on the features of particular parallel systems, is required in order to increase the efficiency of methods. The paper proposes a new parallel hybrid mind evolutionary computation algorithm (HMEC), which belongs to a class of memetic algorithms for solving global optimization problems. The distinct feature of this algorithm consists in using a set of different memes, which allows the algorithm to adapt to various objective functions. In this context, a meme represents any local optimization method, which improves a current solution at particular stages of the main algorithm. Generally, memetic algorithms are hybridization of a population method and one or several local optimization methods. The proposed algorithm is designed for the loosely coupled computational systems made of desktop computers. Software implementation of the algorithm, which includes an operation of decomposing the search domain at the stage of the first initialization of population, was developed in this work. As the main implementation tools, Python programming language and data exchange interface MPI were utilized. MPI is known to be one of the most common interfaces for loosely coupled systems. An extensive investigation of the algorithm performance was conducted using the modified benchmark optimization functions: Rastrigin, Rosenbrock, Weierstrass, and Griewank. The efficiency of HMEC was estimated by the probability of global optimum localization, the average number of iterations and the average number of trials. The number of victories for each meme during local competitions in the “multi-start” mode was also measured and analyzed. Results showed that the parallel implementation of HMEC is more capable of finding a high-quality minimum of an objective function (in terms of both probability and accuracy) than the sequential one. Modifications of the parallel hybrid algorithm as well as the whole multi-memetic approach proved to be promising and are worth of further investigation. References
Publications with keywords: distributed computing, global optimization, Mind Evolutionary Computation algorithm, hybrid algorithms, evolutionary computation Publications with words: distributed computing, global optimization, Mind Evolutionary Computation algorithm, hybrid algorithms, evolutionary computation See also: Thematic rubrics: Поделиться:
|
|
|||||||||||||||||||||||||||||||
|