A multi-parent polynomial sampling framework for steady-state real-coded genetic algorithms

doi:10.1016/j.swevo.2026.102419

A multi-parent polynomial sampling framework for steady-state real-coded genetic algorithms

2026-05-24

0

PRE

AI

M

M. M. Raghuwanshi *

P

Purushottam D. Shobhane

DOI：10.1016/j.swevo.2026.102419

Original

Original request for help

Save

Abstract

En 中文

This study proposes a Multi-Parent Steady-State Real-Coded Genetic Algorithm (MP-SS-RCGA) framework to address the fundamental challenge of balancing exploration and exploitation in continuous optimization. This work introduces two novel recombination operators: Multi-Parent Polynomial Mean-Centric Crossover (MPCX), which promotes diversity through centroid-based sampling, and Multi-Parent Polynomial Parent-Centric Crossover (MPPX), which enhances exploitation by sampling around the elite parent. Both operators utilize inter-parent dispersion to induce an implicit, self-adaptive step-size mechanism without requiring external parameter control. We conduct a comprehensive experimental evaluation on a diverse set of unimodal and multimodal benchmark functions, including scalability analysis up to 500 dimensions. To ensure rigorous benchmarking and eliminate coordinate-axis bias, the framework is further evaluated on the CEC-2017 and CEC-2021 competition suites featuring shifted and rotated landscapes. Empirical results demonstrate that MPPX achieves superior convergence accuracy, reduced variance, and strong robustness across problem classes, while maintaining low sensitivity to algorithmic parameters. Comparative analysis with classical (SBX, PCX) and recent crossover operators (PSOX, LogX) confirms that the observed performance gains arise from the proposed geometry-driven sampling mechanism with dispersion-based implicit step-size adaptation. Statistical analysis using non-parametric tests further validates the significance of these improvements. In addition, validation on constrained engineering design problems demonstrates practical applicability and stability. The findings highlight the critical role of offspring reference geometry in governing search dynamics and establish multi-parent polynomial sampling as a scalable, rotation-invariant, and computationally efficient recombination strategy for complex continuous optimization.

Keywords:

Multi-Parent Polynomial Sampling

Steady-State Genetic Algorithm

Continuous Optimization

Exploration-Exploitation Balance

Crossover Operators

A multi-parent polynomial sampling framework for steady-state real-coded genetic algorithms

Abstract

Journal

Organization

Cited Papers