CAJ | 학술논문

根据自适应离散差分进化(SaDDE)算法的提出过程,对算法策略选择问题进行了重点研究。策略池在SaDDE中起着重要作用,策略池的设计面临着3个问题,即：(1)怎样鉴别某个候选解产生策略(CSGS)是有效的还是无效的；(2)应该选择哪些 CSGS 组成策略池；(3)策略池的大小应该是多少。为了解决这些问题,提出了基于相对排列顺序的标度法(RPOSM)和基于 RPOSM 的层次分析法(RPOSM-AHP)。主要采用某电子对抗(electronic countermeasure,简称ECM)仿真实验平台上的6个测试实例(T_INS)进行测试实验。首先,设计了144个不同的CSGS,为了获得这些CSGS在求解问题上的性能排序序列,做了144×6个独立的实验；然后,采用RPOSM和RPOSM-AHP计算这144个CSGS的最终优先级向量；接着,设计了16个具有不同策略池大小的算法,然后在同样的6个测试实例上测试这些算法的性能；最后,再一次采用RPOSM和RPOSM-AHP为SaDDE寻找到了合适的策略池大小。与其他类似算法的对比实验结果表明：在有限的评估次数(NFE)内,SaDDE比同类算法性能优越。
근거자괄응리산차분진화(SaDDE)산법적제출과정,대산법책략선택문제진행료중점연구。책략지재SaDDE중기착중요작용,책략지적설계면림착3개문제,즉：(1)즘양감별모개후선해산생책략(CSGS)시유효적환시무효적；(2)응해선택나사 CSGS 조성책략지；(3)책략지적대소응해시다소。위료해결저사문제,제출료기우상대배렬순서적표도법(RPOSM)화기우 RPOSM 적층차분석법(RPOSM-AHP)。주요채용모전자대항(electronic countermeasure,간칭ECM)방진실험평태상적6개측시실례(T_INS)진행측시실험。수선,설계료144개불동적CSGS,위료획득저사CSGS재구해문제상적성능배서서렬,주료144×6개독립적실험；연후,채용RPOSM화RPOSM-AHP계산저144개CSGS적최종우선급향량；접착,설계료16개구유불동책략지대소적산법,연후재동양적6개측시실례상측시저사산법적성능；최후,재일차채용RPOSM화RPOSM-AHP위SaDDE심조도료합괄적책략지대소。여기타유사산법적대비실험결과표명：재유한적평고차수(NFE)내,SaDDE비동류산법성능우월。
In line with the proposing process of the self-adaptive discrete differential evolution (SaDDE) algorithm, this research focuses on the strategy selection problem. The strategy pool plays a significant role in the SaDDE algorithm, and there are three issues need to be addressed in designing the strategy pool:(1) how to determine if a candidate solution generating strategy (CSGS) is effective;(2) which CSGSes to choose to constitute the strategy pool;and (3) how to find a suitable size forthe strategy pool. In order to solve these problems, a relative permutation order based scale method (RPOSM) and a RPOSM based analytic hierarchy process (RPOSM-AHP) are proposed in this paper. The experiments are mainly conducted on six test instances (T_INSes) which come from an electronic countermeasure (ECM) simulation experimental platform. 144 different CSGSes are designed, and 144×6 independent experiments are performed to obtain the sort sequences of the CSGSes. The RPOSM and the RPOSM-AHP are adopted to obtain the priority vector of the 144 CSGSes. Sequentially, 16 algorithms with different sizes of strategy pools are constructed and their performance is tested on the six T_INSes. Further, the RPOSM and RPOSM-AHP are employed again to find the suitable pool size for the SaDDE algorithm. Computational comparisons demonstrate that, within fixed number of fitness evaluations (NFE), the SaDDE algorithm can generate better results than its competitors.