基于可解释预测模型的Z型立柱片剪切刚度人工认知

Artificial cognition for shear stiffness of upright frames in Z-pattern based on interpretable prediction mode

由立柱、斜横撑等钢结构件组合而成的立柱片的剪切刚度对高层工业货架侧向抗震性能至关重要。由于结构件的特殊性以及连接方式的多样性,目前使用通用的解析计算模型描述立柱片的剪切行为还比较困难。为此,以Z型立柱片为例提出了一种可解释的立柱片剪切刚度预测模型。通过基于性能试验的有限元仿真获取了不同立柱片结构的剪切刚度数据,并基于XGBoost算法建立立柱片剪切刚度可解释预测模型。与神经网络等其他机器学习算法相比,该预测模型分析与有限元数值计算结果接近且便于理解。以此为基础,采用SHAP(SHapley addictive explanation)方法以可视化的方式对Z型立柱片剪切行为进行人工认知。结果显示,提高剪切性能的关键在于立柱和斜横撑截面的设计以及斜横撑结构配置模式;而不同结构配置模式下,可以采用针对性的设计方法对立柱片剪切性能进行改进。

The shear stiffness of upright frame,composing uprights and bracing members,is critical to the seismic performance of high-rise industrial racks in the cross-aisle direction.Given the specificity of structural members and the diversity of upright frame connections,it is challenging to accurately describe shear behavior using a the general analytical model.An interpretable prediction model for shear stiffness of upright frame in Z-Pattern is proposed.The shear stiffness data of different upright frames were obtained by the finite element simulation based on mechanical tests,and the prediction model based on XGBoost had been established.In comparison to other machine learning algorithms,including artificial neural networks,the predictions generated by the established model exhibited greater proximity to the finite element simulation results,demonstrating its high degree of interpretability and understandability.On this foundation,the SHAP method was employed to achieve the artificial cognitive about the shear behavior of upright frame in a visual way.The analysis revealed that the critical factors influencing shear performance were the design of the cross-section of the upright and bracing members,as well as the bracing configuration.The specific structural design methods could be tailored to enhance the shear performance of upright frames,depending on the varying bracing configuration modes.