Parallel hybrid genetic algorithms on consumer-level graphics hardware

In this paper, we report a parallel Hybrid Genetic Algorithm (HGA) on consumer-level graphics cards. HGA extends the classical genetic algorithm by incorporating the Cauchy mutation operator from evolutionary programming. In our parallel HGA, all steps except the random number generation procedure are performed in Graphics Processing Unit (GPU) and thus our parallel HGA can be executed effectively and efficiently. We propose the pseudo-deterministic selection method which is comparable to the traditional global selection approach with significant execution time performance advantages. We perform experiments to compare our parallel HGA with our previous parallel FEP (Fast Evolutionary programming) and demonstrate that the former is much more effective and efficient than the latter. The parallel and sequential implementations of HGA are compared in a number of experiments, it is observed that the former outperforms the latter significantly. The effectiveness and efficiency of the pseudodeterministic selection method is also studied