Serial DPGA vs. Parallel Multithreaded DPGA: Threading Aspects

Abstract

The multiple main populations, reserve populations and subpopulations concepts of a Genetic Algorithms (GAs) offers the advantage of diversity. However, as the population evolves, the GA loses its diversity. As the population converges, it begins to lose its diversity and cannot avoid the local optima problem. This problem is known as Premature Convergence for Parallel GAs (PGA) too. The paper compares the Binary encoded Simple GA (SGA), Binary encoded Serial/ Sequential Dual Population Genetic Algorithm (SDPGA) and Binary encoded Multithreaded Parallel DPGA (MPDPGA) performances for function optimization on multicore system. The Dual Population Genetic Algorithm (DPGA) is an evolutionary algorithm that uses an extra population called the reserve population to provide additional diversity to the main population through crossbreeding. The experimental results on unimodal and multimodal classes of test problem shows the MPDPGA outperforms over SGA and SDPGA. The performance of MPDPGA with DPGA1 is better in terms of accuracy, number of generations and execution time on multicore system. The performance of MPDPGA with DPGA-ED1 is better for Rosenbrock and Schwefel whereas worse for Ackley and Griewangk.KeywordsSerial/Sequential Dual-Population GAParallel Dual-Population GAFunction OptimizationPremature ConvergenceDiversity