| The requirement of the computation speed, the system reliability and the cost effective will urge to develop new computer module, to replace the traditional Von Neumann's structure computer. As the rapid development of network technology, it's possible to realize parallel and distributed computing. Thus, the algorithm research which is suitable for parallel computing has become the hot spot of the front.In this paper, it brings into focus on Distributed Parallel Evolutionary Algorithms based on MPI. The summary is following: firstly, parallel Genetic Algorithm is discussed, the experiments results the algorithm time complexity is reducing; Secondly, Simulated Annealing is discussed, the performance of algorithm scales up with the increase of processors, and the character of asynchronous parallel make algorithm more suitable for all kinds of processor; Thirdly, after analyzing hybrid optimization strategy of genetic algorithm and simulated annealing algorithm, a parallel simulated annealing genetic based on MPI is developed, aiming at the deficiency of the Simulated Annealing Genetic Algorithm(SA-GA), the idea of parallel evolution was combined into SA-GA and the parallel SA-GA algorithm was proponed based on MPI. The task assignments and communication overheads of this algorithm has been analyzed. Through emulational experiment by test function, it has indicated that the parallel algorithm has improved the running speed and astringency and it is easier to find a global optimal solution, which possesses expansibility and acquires linear accelerate rate; Finally, a new parallel optimization method based on niching hybrid genetic simulated annealing algorithm is proposed in this paper. And the new method's features are discussed. Then it is applied to the Shubert function, a representative multi-model optimization problem. The experimental results verify that the proposed parallel method improves the convergence efficiency, enhances the ability of global searching. |
PDF Full Text Request