Proposal of a nonparametric optimization method based on evolutionary process evaluation through differentiation (Validation of benchmark function and hysteresis curve identification)

Abstract

Recent advancements in digitalization have resulted in the daily collection of vast amounts of data. To capitalize on this wealth of information, it is imperative to address multivariate issues, many of which are classified as NP-hard problems. One potential solution lies in metaheuristic optimization methods, which offer shorter search times and can generate approximate solutions. These techniques have seen applications across various domains. Nevertheless, a significant challenge posed by numerous representative metaheuristic methods involves the necessity for parameter configurations, the values of which notably impact convergence accuracy. This study proposes a novel optimization methodology grounded in metaheuristic optimization techniques that eliminate the need for problem-dependent accuracy affecting parameter settings. The authors assessed the efficacy of our method using standard benchmark functions and engineering benchmark problems. Furthermore, we employed it to search for multiple variables, such as historical curves, while conducting a nonlinear seismic response analysis in a real-world application scenario. Our findings confirm that our approach is not only more cost-effective but also superior in accuracy compared to previously used metaheuristic optimization methods.