Multi-fidelity meta-optimization for nature inspired optimization algorithms
Hui Li,Zhiguo Huang,Xiao Liu,Chenbo Zeng,Peng Zou
DOI: https://doi.org/10.1016/j.asoc.2020.106619
IF: 8.7
2020-11-01
Applied Soft Computing
Abstract:<p>The last couple of decades have witnessed a steadily increasing applications of nature inspired optimization (NIO) in vast fields such as power engineering, environmental engineering, and civil engineering. Behavioral parameters deeply affect the optimization performance for a NIO algorithm. Meta-optimization is good choice for parameter optimization of NIOs, which uses an optimizer to optimize another optimizer. However, meta-optimization is a very time-consuming process. For this reason, we propose a multi-fidelity strategy based meta-optimization approach to speed up the parameter optimization. Four types of fidelity control functions determine the fidelity level in the course of meta-optimization. We test the proposed method in the meta-optimization systems with diverse meta-NIOs (cuckoo search, fruit fly optimizer, gray wolf optimizer, krill herd, and whale optimization algorithm), diverse optimized-NIOs (cuckoo search, differential evolution, particle swarm optimizer, squirrel search algorithm, and water wave optimizer), and diverse benchmark problems (Ackley-50, Eggholder-2, Michalewicz-5, Shubert-2, Sphere-50, and F1–20). We also apply it to a real-world engineering problem to estimate the source terms of gas emission. Experimental results indicate that multi-fidelity strategy can substantially speed up meta-optimization systems and hence has the potential to be generalized to various NIOs.</p>
computer science, artificial intelligence, interdisciplinary applications