A Study On The Multi-population Parallel Evolutionary Algorithms

Artificial bee colony algorithm is a new high-efficiency global optimization algorithm, but it is easier to fall into local optimal with serial single population. Moreover, due to its greedy selection approach, it has slow convergence rate at the later evolution stage. In order to overcome these shortcomings, This thesis introduces the cross entropy method combined with artificial swarm algorithm, proposes a new cross entropy colony algorithm(CEABC). Parallel evolutionary algorithm is also the direction of optimization algorithm development in recent years, however, the current parallel algorithms has little improvement in calculation and stability, and the research of parallel artificial colony algorithm is only at the early stage. Therefore, this thesis proposes coarse-grained parallel artificial colony algorithm(CPABC) and coarse-grained parallel cross entropy colony algorithm(CPCEABC) from the perspective of multi-populations which is based on principle of the ABC algorithm characteristics.Cross entropy method can successfully combine the onlooker generation mechanism of ABC, instead of roulette wheel selection approach. CPABC and CPCEABC combine the scout bee mechanism of ABC in the structure, strengthen the ability to avoid premature convergence. Experimental results which based on benchmark functions indicate that three new algorithms on optimization ability and convergence speed are improved obviously. Then this thesis disperses CPABC to apply in TSP problems. The new algorithm can find the best path for both examples, the best results are better than serial algorithm. Then two parallel algorithms are tested with four Antenna optimization benchmark problems, all results are better than ABC. Finally, two algorithms are applied to antenna array pattern synthesis, getting low side-lobe level.New parallel algorithms balance global optimization and convergence speed, have strong Versatility, good stability, small amount of calculation, they are more suitable for applying in high complexity engineering problems.