Online Cluster-Based Parameter Control for Metaheuristic
Vasileios A. Tatsis, Dimos Ioannidis
TL;DR
The paper addresses the challenge of robust online parameter control for metaheuristics by proposing Cluster-Based Parameter Adaptation (CPA), a three-stage framework that uses K-means clustering to identify promising regions in the parameter search space and to generate new parameter values around those regions. CPA operates in an online, population-based setting with Stage 0 random exploration, Stage 1 guided generation around found clusters using evaporation-based distance sampling, and Stage 2 dynamic deployment of the generated parameters into the underlying optimizer, here Differential Evolution (DE). Empirical evaluation on the SOCO 2011 and CEC 2013 test suites shows CPA-DE achieves competitive and often superior performance compared to state-of-the-art adaptive methods like SHADE, with robustness across high-dimensional problems. The findings suggest that integrating unsupervised learning for parameter control is a promising direction for enhancing real-time optimization and motivates future hybrid approaches combining machine learning with evolutionary search mechanisms.
Abstract
The concept of parameter setting is a crucial and significant process in metaheuristics since it can majorly impact their performance. It is a highly complex and challenging procedure since it requires a deep understanding of the optimization algorithm and the optimization problem at hand. In recent years, the upcoming rise of autonomous decision systems has attracted ongoing scientific interest in this direction, utilizing a considerable number of parameter-tuning methods. There are two types of methods: offline and online. Online methods usually excel in complex real-world problems, as they can offer dynamic parameter control throughout the execution of the algorithm. The present work proposes a general-purpose online parameter-tuning method called Cluster-Based Parameter Adaptation (CPA) for population-based metaheuristics. The main idea lies in the identification of promising areas within the parameter search space and in the generation of new parameters around these areas. The method's validity has been demonstrated using the differential evolution algorithm and verified in established test suites of low- and high-dimensional problems. The obtained results are statistically analyzed and compared with state-of-the-art algorithms, including advanced auto-tuning approaches. The analysis reveals the promising solid CPA's performance as well as its robustness under a variety of benchmark problems and dimensions.