Study on the application of discrepancy-guided symbolic regression algorithm in analyzing the impact resistance of UHP-SFRC target against high velocity projectile impact

Created by W.Langdon from gp-bibliography.bib Revision:1.8414

@Article{Zou:2025:ijimpeng,

• author = "Delei Zou and Dilyar Thoti and Zhihui Bao",
• title = "Study on the application of discrepancy-guided symbolic regression algorithm in analyzing the impact resistance of {UHP-SFRC} target against high velocity projectile impact",
• journal = "International Journal of Impact Engineering",
• year = "2025",
• volume = "201",
• pages = "105276",
• keywords = "genetic algorithms, genetic programming, UHP-SFRC targets, High-velocity projectile impacts, Empirical formulas, DOP, Symbolic regression, Discrepancy-guided",
• ISSN = "0734-743X",
• URL = "https://www.sciencedirect.com/science/article/pii/S0734743X25000570",
• DOI = "doi:10.1016/j.ijimpeng.2025.105276",
• abstract = "Accurately predicting the depth of penetration (DOP) values for ultra-high-performance steel fiber-reinforced concrete (UHP-SFRC) targets under high-velocity projectile impact (HVPI) is crucial to assess damage patterns and maintain the integrity of protective structures. However, traditional empirical formulas often fall short, resulting in significant underestimations. To address these inaccuracies and inefficiencies, this study establishes a discrepancy-guided, H2O Auto-ML-assisted Offspring Selection Genetic Programming-Symbolic Regression (OSGP-SR) algorithm to rectify traditional empirical formulas and markedly improve the accuracy and robustness of DOP predictions. Four widely used empirical formulations for predicting DOP in traditional reinforced concrete structures, i.e., BRL, NDRC, ACE, and AW were selected as optimisation objectives. A dataset comprising 265 samples of UHP-SFRC targets subjected to HVPI was created for verification and analysis. An examination of the prediction deviations in the original formulations informed an optimisation strategy based on discrepancy-guided analysis. Moreover, the identification of key feature variables was conducted through feature contribution screening, and the proposed algorithm based on discrepancy-guided was used to refine the existing formulas. Results indicate that the modified NDRC model achieves the lowest mean absolute percentage error (MAPE), while the BRL model excels in comprehensive evaluation, maintaining accuracy regardless of coarse aggregate presence. Furthermore, a graphical user interface (GUI) for engineering applications is provided",

}

Genetic Programming entries for Delei Zou Dilyar Thoti Zhihui Bao

Citations