PEAB: A pool-based distributed evolutionary algorithm model with buffer

Pool Model is an asynchronous, loosely coupled distributed evolutionary algorithm (dEA) design architecture. However, the classical Pool Model face some design problems, such as population control, work redundancy, rough selection/replacement strategies, and unreliable connections, etc. In this paper, a novel distributed pool evolutionary algorithm (EA) model with buffer (PEAB) is proposed. PEAB can solve the inherent problems of the Pool Model by using the buffer setting, the Reunion mechanism, and the Migration in Pool (MP) strategy. Besides, PEAB provides stronger population control and more global population selection/replacement strategies. In the experimental part, we compared PEAB with another Pool Model named EvoSpace using a common benchmark. The experiments showed that the convergence rate of PEAB is 59.7% faster than that of EvoSpace under the respective fastest conditions. PEAB also has a faster reception rate of the first generation and stronger population control. Besides, this paper also tests and analyzes the scalability of PEAB using two other benchmarks. The overall trend of the experiment results suggested that PEAB would be faster with more Workers. Last but not least, this paper studies the effect of the MP strategy on the performance of PEAB, and the results showed that the MP strategy can effectively improve the convergence efficiency.

Automated summary

The paper introduces a novel distributed pool evolutionary algorithm model PEAB, which addresses the issues of the classical Pool Model through buffer setting, Reunion mechanism, and MP strategy. Experimental results demonstrate that PEAB has a faster convergence rate and stronger population control compared to EvoSpace.