Component-based design of multi-objective evolutionary algorithms using the Tigon optimization library
Component-based design of multi-objective evolutionary algorithms using the Tigon optimization library
复制标题
DOI：
10.1145/3449726.3463194
复制
发表时间：
2021-07
期刊：
影响因子：
--
通讯作者：
João A. Duro;Daniel C. Oara;Ambuj Sriwastava;Yiming Yan;Shaul Salomon;R. Purshouse
中科院分区：
文献类型：
--
作者：
João A. Duro;Daniel C. Oara;Ambuj Sriwastava;Yiming Yan;Shaul Salomon;R. Purshouse
多目标优化问题涉及多个相互冲突的目标，必须同时进行优化。生成完整的帕累托最优前沿 (POF) 的计算成本可能很高，甚至不可行，因此人们对使用多目标进化算法 (MOEA) 产生了极大的兴趣，众所周知，这种算法可以生成 POF 的良好近似值。 MOEA 可能很难实施，即使对于经验丰富的优化专家来说，这也可能是一项非常耗时的任务。因此，文献中存在多个优化库，提供对最流行的 MOEA 的现成访问。一些优化库还提供了设计 MOEA 的框架。然而，现有框架可能过于严格，无法为更复杂的 MOEA 的设计提供足够的灵活性。为了解决这个问题，最近提出了一个称为 Tigon 的优化库，它采用基于组件的 MOEA 设计框架，重点关注灵活性和可重用性。本文通过展示如何实现不同类型的 MOEA 来展示这个新框架的通用性，涵盖了进化计算中的几种范式。本文中的工作可以指导研究人员和其他类似人员使用 Tigon 优化库构建自己的 MOEA。
Multi-objective optimization problems involve several conflicting objectives that have to be optimized simultaneously. Generating a complete Pareto-optimal front (POF) can be computationally expensive or even infeasible, and for that reason there has been an enormous interest in using multi-objective evolutionary algorithms (MOEAs), which are known to generate a good approximation of the POF. MOEAs can be difficult to implement, and even for experienced optimization experts it can be a very time consuming task. For this reason several optimization libraries exist in the literature, providing off-the-shelf access to the most popular MOEAs. Some optimization libraries also provide a framework to design MOEAs. However, existing frameworks can be too stringent and do not provide sufficient flexibility for the design of more sophisticated MOEAs. To address this, a recently proposed optimization library, known as Tigon, features a component-based framework for the design of MOEAs with a focus on flexibility and re-usability. This paper demonstrates the generality of this new framework by showing how to implement different types of MOEAs, covering several paradigms in evolutionary computation. The work in this paper serves as a guide for researchers, and others alike, to build their own MOEAs by using the Tigon optimization library.