Efficient Evolutionary Algorithms for Many-objective Optimization

多目标优化的高效进化算法

• 批准号: RGPIN-2015-03651
• 负责人: Rahnamayan, Shahryar
• 金额: $ 2.04万
• 依托单位: University of Ontario Institute of Technology
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2020
• 资助国家: 加拿大
• 起止时间: 2020-01-01 至 2021-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=708145
• 关键词: Efficient; Evolutionary; Algorithms; Many; objective

Most of the time, when we are talking about an improvement of something, in fact, we are interested to minimize or maximize (i.e. optimize) quality or quantity of an entity. Universal examples are maximization of reliability, efficiency, safety, and benefit; or minimization of the pollution, risk, consumed energy, or production time/cost. Now, it is clear why the fingerprint of the optimization is visible in all science and engineering fields, ranging from healthcare to astronomy. In this direction, nature-inspired problem solving methods play a crucial role to efficiently solve complex real-world problems. Evolutionary Algorithms (EAs) are well-known examples inspired from the genetic biology; they employ biological operations such as selection, crossover, and mutation. EAs are pioneers tackling problems which are hard or even impossible to be solved by the conventional methods. For majority of our practical problems, we are faced with two or more (multi) conflicting objectives to optimize simultaneously; such as minimizing cost and maximizing efficiency for a system. The current successful evolutionary multi-objective algorithms have focused on problems with two or three objectives. However, recently, we face with problems which consist of more than three objectives (called many-objective). EAs have demonstrated their niche in solving these problems due to the requirement of finding multiple trade-off solutions for these problems. However, having a number of algorithmic restrictions, these methods were shown to be non-scalable to many-objective problems. These kinds of problems present new challenges for algorithm design and visualization which have not been addressed properly. This research program expects training 3 PhD and 3 MSc students by involving them in the cutting-edge research topics. These topics address the existing restrictions by enhancing various correlated aspects, namely, a) designing computationally fast algorithms, b) utilizing decomposition methods for dividing an conquering the original problem, c) designing tailored processing type (i.e., sequential, distributed, and parallel), d) designing simple and intuitive large-scale data visualization techniques (for better understanding and supporting an interactive computation), and e) designing effective performance metrics. The outcomes of the current research will be beneficial for a wide range of research communities and industrial sectors in Canada which utilize optimization by any means in scheduling, control systems, robotics, data mining, circuits design, communications, bioinformatics, image processing, networking, traffic engineering, etc. The applicant's more than ten years' comprehensive experience in evolutionary computation will play a pivotal role in success of this research program.

大多数时候，当我们谈论某事物的改进时，事实上，我们感兴趣的是最小化或最大化（即优化）实体的质量或数量。普遍的例子是可靠性、效率、安全性和效益的最大化；或最小化污染、风险、消耗的能源或生产时间/成本。现在，很清楚为什么优化的痕迹在从医疗保健到天文学的所有科学和工程领域中都可见。在这个方向上，受自然启发的问题解决方法在有效解决复杂的现实世界问题方面发挥着至关重要的作用。进化算法 (EA) 是受遗传生物学启发的著名例子；它们采用选择、交叉和突变等生物操作。 EA 是解决传统方法难以甚至不可能解决的问题的先驱。对于大多数实际问题，我们面临着两个或多个（多个）相互冲突的目标同时优化；例如最小化系统成本和最大化系统效率。当前成功的进化多目标算法主要关注具有两个或三个目标的问题。然而，最近我们面临着由三个以上目标组成的问题（称为多目标）。由于需要为这些问题找到多种权衡解决方案，EA 已经证明了它们在解决这些问题方面的优势。然而，由于存在许多算法限制，这些方法被证明无法扩展到多目标问题。这类问题给算法设计和可视化带来了新的挑战，但尚未得到妥善解决。该研究项目预计培养3名博士生和3名硕士生，让他们参与前沿研究课题。这些主题通过增强各种相关方面来解决现有的限制，即，a）设计计算快速的算法，b）利用分解方法来划分和解决原始问题，c）设计定制的处理类型（即顺序、分布式和并行） ，d）设计简单直观的大规模数据可视化技术（为了更好地理解和支持交互式计算），以及e）设计有效的性能指标。当前研究的成果将有益于加拿大广泛的研究团体和工业部门，他们在调度、控制系统、机器人、数据挖掘、电路设计、通信、生物信息学、图像处理、网络、申请人十余年在进化计算方面的综合经验将对本研究项目的成功起到关键作用。