Development and Analysis of Metaheuristics for Challenging Large-Scale Optimization Problems

点击翻译按钮获取中文摘要

The benefits of optimization can be observed in almost all aspects of life, from science, healthcare, economics, to engineering problems. Important problems in these areas are computationally intractable by exact methods. My research uses computational intelligence, a branch of artificial intelligence to solve problems that are either too challenging or impossible to solve using conventional optimization methods. Specifically, I use algorithms that simulate the emergent behaviour of flocking birds (particle swarm optimization, PSO) and evolutionary algorithms (EAs) inspired by Darwinian evolution and population genetics: I use biological processes such as selection, crossover, and mutation to quickly find solutions that are acceptable in practice, or "near perfect" where the optimal solution may require a lifetime to compute. The majority of real-world optimization problems involve simultaneous optimization of two or more, often conflicting objectives, e.g., minimizing transportation costs while maximizing safety and efficiency; this is called a multi-objective optimization problem (MOP). The successes in the role played by optimizing algorithms based on EAs and PSOs, among other metaheuristics in solving MOPs are well documented. Increasingly we are faced with problems that have more than three objectives, i.e., a special case of MOPs, termed many-objective optimization problems (MaOPs). However, the performance of the aforementioned algorithms deteriorates severely with increase in the number of objectives (beyond three) and the number of problem dimensions or decision variables, i.e., for large-scale many-objective optimization problems, which present new challenges for optimization techniques. The proposed research will tackle efficient algorithm design, solution evaluation, and data/algorithm activity visualizations that have not been adequately addressed for MaOPs. Both well-known benchmark functions for MaOPs and large-scale real-world optimization problems arising in network science, such as finding critical nodes in complex networks, will be investigated. The critical node detection problem has many practical applications: e.g., targeting specific proteins in protein interaction networks for use in drug design, in computer network security so as to protect or attack important areas, in determining smart grid vulnerability, and in identifying individuals for vaccination or quarantine when mitigating disease spread. Providing efficient algorithmic solutions for important problems, and by improving our understanding of appropriate algorithms and techniques for different classes of challenging optimization problems, this research will benefit many researchers and practitioners that utilize optimization across different domains. I will involve graduate and senior undergraduate students to find cutting-edge solutions to challenging optimization problems and to advance the theoretical aspects and development of EAs and PSOs.