Aerodynamic optimization using a parallel asynchronous evolutionary algorithm controlled by strongly interacting demes

摘要

A parallel asynchronous evolutionary algorithm controlled by strongly interacting demes for single- and multi-objective optimization problems is proposed. It is suitable even for non-homogeneous, multiprocessor systems, ensuring maximum exploitation of the available processors. The search algorithm utilizes a structured topology of evaluation agents organized in a number of inter-communicating demes arranged on a 2D supporting mesh. Once an evaluation terminates and a processor becomes idle, a series of intraand inter-deme processes determines the next agent to undergo evaluation on this specific processor. Real coding and differential evolution operators are used. Mathematical and aerodynamic-turbomachinery optimization problems are presented to assess the proposed method in terms of CPU cost, parallel efficiency and quality of solutions obtained within a predefined number of evaluations. Comparisons with conventional evolutionary algorithms, parallelized based on the master-slave model on the same computational platform, are presented.