Research On Hybrid Swarm Intelligence Parallel Algorithm Based On Multi-core Cluster

In recent years, Swarm Intelligence Optimization Algorithm is a new kind ofbionic optimization algorithm, more research about two swarm intelligencealgorithms are Artificial Fish Swarm Algorithm (AFS) and Artificial Bee ColonySwarm Algorithm (ABC). Because of these two algorithms have strong robustness,simple parameter setting, easy to implement and so on. Has extensive research andapplication in evolutionary computation, genetic control optimization strategy, it’sbecome very active frontier topics in interdisciplinary science.As a new evolutionary computation technique, swarm intelligence algorithm, inthe simulation animal sampling, taking a different approach to solving complexproblems in the process of collecting food. AFS and ABC have strong global searchcapability, but due to poor post-fine search capabilities, easy to fall into localoptimum. So, first to improve optimization strategy of AFS and ABC, uniformlydistributed on the initial population, then randomly divided into two populations, anduse interactive learning strategies to accelerate the convergence rate. In preliminaryalgorithm by AFS and ABC, get range optimization. The late adoption of improved byDN-AFS and RP-ABC algorithm to get the solution for local pre-fine search. Finally,build MPI+OpenMP+STM parallel programming model for parallel algorithmdesign and analysis. Through simulation, to prove the feasibility and effectiveness ofthe hybrid swarm intelligence parallel algorithm.The main work includes:(1) Because the weak late-search capability of AFSA, reduced populationdiversity, easy to fall into local optimum. In the search process of artificial fish adoptdynamic weighting factor strategy and dynamically adjust the step size of the field ofview, balancing the global search capability; by constant threshold determine thedegree of aggregation of fish, introduction of niche mechanism, then analysis ofparallel algorithms, put forward a new artificial fish swarm Parallel Dynamic weighNiches Artificial Fish Swarm (PDN-AFS) algorithm. Simulation experiments showthe effectiveness of PDN-AFS algorithm in multi-core cluster environment.(2) By analyzing the characteristics of artificial bee colony algorithm, when itdealing with complex function optimization convergence is slow, easier to fall into the"premature". In the search process of artificial fish adopt random perturbation factor η and global optimal solution gbest two strategies, get the new formula of update thefood source, application OpenMP parallel technology parallel rewritten moretime-consuming algorithms block, put forward a new artificial fish swarm parallelartificial bee colony (PRP-ABC) algorithm. Optimization results show that PRP-ABChas a higher speedup.(3) By analyzing the two improved algorithm of the above, combined with thestrong global optimization ability of AFS and ABC, proposed hybrid swarmintelligence parallel algorithm. Randomly divided into two groups, directly use ABCand AFS for early optimization, locate the global optimal solution, in the lattermethod, the obtained solution is also divided into two groups performed, one groupusing DN-AFS implementation, another using RP-ABC, enhance the ability of localsearch algorithms in the late. Finally, build MPI+OpenMP+STM parallelprogramming model, through the complex function optimization experiments provedthe optimization efficiency of hybrid algorithm.