Seismic performance optimization design of concrete beam-column joints based on NSGA-II multi-objective optimization algorithm

In modern building construction, reinforced concrete beam-column frame structures serve as the primary load-bearing system in high-rise projects, where the performance of individual components critically influences overall structural integrity. Current design practices predominantly rely on empirical code-based data and single-objective safety factor constraints, an approach that often leads to underutilized material properties and design inefficiency. When confronting the high-dimensional, nonlinear optimization problems inherent in seismic performance design, conventional methods exhibit low computational efficiency and frequently yield poorly distributed solution sets, thereby failing to effectively balance competing objectives such as safety, cost, and environmental impact. To address these limitations - specifically low efficiency and uneven solution distribution in high-dimensional nonlinear problems - this study enhances the NSGA-II framework (Non-dominated Sorting Genetic Algorithm II) with Pareto optimality principles. The proposed improvements incorporate an adaptive parameter-tuning strategy and a Kriging surrogate model to improve algorithmic convergence efficiency. Innovatively, this work integrates seismic performance metrics, Life Cycle Cost (LCC), and carbon emissions into a unified multi-objective optimization system to establish a more comprehensive design evaluation criterion. By employing a Deep Neural Network (DNN) to accelerate Finite Element Method (FEM) simulations, computational time for optimization iterations is reduced by over 65%. The results demonstrate that the optimized beam-column designs achieve material cost reductions of 12%-18% while maintaining equivalent seismic fortification levels and satisfying the seismic detailing requirements of ACI 318-19. The proposed multi-objective synergistic optimization approach overcomes the limitations of traditional single-objective design methods.