<![CDATA[This study proposes a hybrid framework that integrates structural analysis, dimensionality reduction (DR), and metaheuristic algorithms to enhance steel structure design optimization. Structural responses are extracted using ETABS, and high-dimensional design data are reduced via PCA, t-SNE, and deep Autoencoders. Classifiers such as XGBoost, LightGBM, and CatBoost are employed as diagnostic tools to verify that the reduced latent space preserves feasibility boundaries. Optimization is then performed using Particle Swarm Optimization (PSO) and Symbiotic Organisms Search (SOS) to minimize structural weight while meeting code requirements. The framework is validated on benchmark functions and a 20-story braced steel frame, showing that SOS consistently outperforms PSO and that MSE-based Autoencoders achieve superior convergence and solution quality compared to linear or shallow methods. The results highlight DR-integrated metaheuristics as a scalable and effective approach for large-scale optimization. Future work will extend the framework to predictive models and more complex structural systems.]]>
Copyrights © 2026
