ABSTRACT

Recently developed Beetle Antennae Search algorithm (BAS) mimics the odor sensing mechanism of the longhorn beetles. The beetles have many species and many of these are advantageous to the nature as well as the mankind. Excepting the odor sensing activity, the beetles are naturally strong insects, and some of them have storage memory for adaptive learning and showcase social behavior. These natural mechanisms make them intelligent enough to perform the routine tasks for existence. This article proposes a novel Storage (Memory) Adaptive Collaborative BAS (SACBAS) algorithm, which incorporates the memory stored adaptive learning. This helps exploit the Group Extreme Value (GEV) instead of the Individual Extreme Values in swarm for faster convergence. Further, the SACBAS uses the reference points based on non-dominated sorting to diversify the state space. To test the data-driven performance of SACBAS, the Support Vector Machine (SVM) algorithm with linear kernel is used in this study. First, the SACBAS algorithm is tested on the multi-objective ZDT and DTLZ test-suites and compared with two recent techniques, the reference points based Non-dominated Sorting Genetic Algorithm (NSGA III) and Multi-Objective Evolutionary Algorithm based on Decomposition (MOEA/D). Second, the data-driven SACBAS is tested with real-world cases based on offline data. The proposed SACBAS is shown to handle the offline data efficiently and obtains promising results. The Friedman Test is carried out to differentiate the SACBAS from other two techniques and the Post Hoc Test confirms that the SACBAS obtains better HyperVolume indicator scores and outperforms the NSGA III and MOEA/D.

摘要

最近开发的甲壳虫天线搜索算法（BAS）模仿了长角甲虫的气味感应机制。甲虫有许多种，其中许多对自然和人类都有好处。除气味感应活动外，甲虫是天然强壮的昆虫，其中一些具有用于适应性学习和展示社交行为的存储记忆。这些自然机制使它们足够智能，可以执行日常任务以维持生存。本文提出了一种新颖的存储（内存）自适应协作BAS（SACBAS）算法，该算法结合了内存存储的自适应学习。这有助于利用群体极值（GEV）而不是群体中的个体极值，从而实现更快的收敛。更多，SACBAS使用基于非支配排序的参考点来分散状态空间。为了测试SACBAS的数据驱动性能，本研究使用具有线性核的支持向量机（SVM）算法。首先，在多目标ZDT和DTLZ测试套件上测试了SACBAS算法，并与两种最新技术进行了比较：基于参考点的非支配排序遗传算法（NSGA III）和基于分解的多目标进化算法（MOEA） / D）。其次，基于离线数据的真实案例对数据驱动的SACBAS进行了测试。所提出的SACBAS被证明可以有效地处理离线数据并获得可喜的结果。