Multi-objective optimization for nonlinear steel frames using a parameter-less MOO algorithm and XGBoost

Abstract

This study proposes a hybrid RDMO-XGBoost method for bi-objective optimization of nonlinear steel frames, targeting minimal structural mass and top-story displacement under static loads. Traditional linear elastic mod- els fail to capture the nonlinear inelastic behavior of steel frames, such as yielding and buckling, necessitating advanced optimization strategies. The Rao-DE Multi-Objective Optimization (RDMO) algorithm, combined with Differential Evolution, ensures diverse and accurate Pareto solutions without parameter tuning. Integrated with XGBoost, a machine learning tool, the method predicts frame responses rapidly, reducing reliance on time-intensive finite element analysis. Applied to a two-story frame, RDMO-XGBoost reduced computation time by over 55% (from 50,400 to 22,370 seconds) compared to generalized differential evolution 3 (GDE3) and RDMO, while maintaining comparable performance in convergence, coverage, and diversity. Anchor point results confirm its effectiveness, achieving lower mass (5148.891 kg) and displacement (0.807 mm), demon- strating its potential for efficient and robust steel frame design optimization.