Parallel Scalable Algorithms with Mixed Local-Global Strategy for Global Optimization Problems
- 539 Downloads
This paper continues development of information-statistical approach to minimization of multiextremal functions in the case of non-convex constraints. Proposed approach is called index method. Solving multidimensional problem is reduced to solving equivalent single dimensional one. Dimension reduction is based on Peano curves that allow mapping multidimensional hyper cube onto the segment on real axis. We also use rotating Peano curves that allowed effectively parallelize algorithm to use hundreds of processors. Special attention was paid to mixed local-global strategy for algorithm convergence acceleration.
Keywordsglobal optimization parallel computing index method local-global strategy
Unable to display preview. Download preview PDF.
- 1.Strongin, R.G.: Global Optimum Search. M.: Znanie (1990)Google Scholar
- 3.Strongin, R.G., Barkalov, K.A.: About convergence of index method in problems of conditional optimizations with ε-reserving solutions Mathematical issues of cybernetics. M.: Nauka, pp. 273–288 (1999)Google Scholar
- 6.Barkalov, K.A.: Convergence acceleration for constrained global optimization problems. Printed Nizhni Novgorod State University, Nizhni Novgorod (2005)Google Scholar
- 7.Barkalov, K.A., Ryabov, V.V., Sidorov, S.V.: Using Peano curves in parallel global optimization. In: Materials of 9th International Conference-Seminal “High-Performance Computing on Cluster Systems”, Vladimir, pp. 44–47 (2009)Google Scholar
- 8.Strongin, R.G., Gergel, V.P., Barkalov, K.A.: Parallel methods of global optimization problems solving. Priborostroenie 52(10), 25–32 (2009)Google Scholar
- 10.Gablonsky, M.J.: Modifications of the DIRECT Algorithm. Ph.D. thesis, North Carolina State University, Raleigh, NC (2001)Google Scholar