建筑钢结构进展2025，Vol.27Issue(12)：1-9,9.DOI:10.13969/j.jzgjgjz.20240628002
基于可解释机器学习的螺栓抗剪连接件承载力预测模型
Prediction Model for the Bearing Capacity of Bolted Shear Connectors Based on Interpretable Machine Learning
摘要
Abstract
Accurate prediction of the shear capacity of bolted shear connectors has significant importance for the design and application of reusable steel-concrete composite beams.This study constructed a database containing 200 sets of experimental results from push-out tests on the shear capacity of bolted shear connectors,through the mining of openly accessible literature resources.During the data processing stage,the study employed interpretable machine learning methods to precisely identify abnormal samples and thoroughly analyze the correlations within the dataset.In the process of model development,key features such as concrete strength,elastic modulus,bolt tensile strength,bolt yield strength,bolt diameter,and embedded length of the push-out test specimens were selected.Based on four advanced machine learning algorithms and a semi-empirical hyperparameter tuning approach,a prediction model for shear capacity was established.After comparison,the optimal machine learning model was identified,achieving a determination coefficient of 0.987 on the test set,significantly surpassing existing empirical models.To further enhance the credibility and transparency of the prediction model,the study employed the additive explanation method of SHAP to interpret the predictions of the optimal machine learning model,revealing the inherent logic and basis of the model's predictions.Additionally,this study innovatively proposed a method for generating dimensionally balanced formulas based on symbolic regression and genetic algorithms.Through this method,a prediction formula for the shear capacity of bolted shear connectors was fitted.This formula exhibits high prediction accuracy,achieving a determination coefficient of 0.899 on the test set,and clearly reveals the explicit mapping relationship between the key features of bolted shear connectors and their shear capacity,providing strong support for practical applications.关键词
抗剪承载力/可解释机器学习/螺栓抗剪连接件/数据驱动/预测模型/可决系数/显性映射关系Key words
shear bearing capacity/interpretable machine learning/bolted shear connector/data-driven/prediction model/determination coefficient/explicit mapping relationship分类
建筑与水利引用本文复制引用
马晨,罗云标..基于可解释机器学习的螺栓抗剪连接件承载力预测模型[J].建筑钢结构进展,2025,27(12):1-9,9.基金项目
国家自然科学基金(52238012,51508381) （52238012,51508381）
