基于可解释机器学习方法的RC深受弯构件开裂剪力预测  

Prediction of the Shear Cracking Strength of RC Deep Beams Based on the Interpretable Machine Learning Approach

作　　者：马财龙[1,2] 王文虎 侯宪龙 谢晨曦 鲁成凤 MA Cailong;WANG Wenhu;HOU Xianlong;XIE Chenxi;LU Chengfeng(School of Civil Engineering and Architecture,Xinjiang University,Urumqi Xinjiang 830017,China;Xinjiang Key Laboratory of Building Structure and Seismic Resistance,Xinjiang University,Urumqi Xinjiang 830017,China)

机构地区：[1]新疆大学建筑工程学院,新疆乌鲁木齐830017 [2]新疆大学新疆建筑结构与抗震重点实验室,新疆乌鲁木齐830017

出　　处：《新疆大学学报（自然科学版）（中英文）》2023年第5期621-629,共9页Journal of Xinjiang University(Natural Science Edition in Chinese and English)

基　　金：新疆天山青年计划-青年博士科技人才培养项目“钢筋混凝土深受弯构件的受剪性能状态划分及机理模型”(2020Q069);新疆大学引进博士科研启动项目“钢筋混凝土框架节点基于性能的受剪设计方法”(620312396).

摘　　要：钢筋混凝土(Reinforced Concrete,RC)深受弯构件易发生脆性剪切破坏,斜裂缝产生及发展伴随全过程,开裂剪力是关键指标之一.建立了276根含开裂荷载信息的RC深受弯构件受剪试验数据库,采用机器学习XGBoost集成算法预测其开裂剪力,并采用5项统计指标评估机器学习模型的预测性能.从无腹筋和双向腹筋两类工况,对比了所建机器学习模型与5个半经验半理论计算公式的预测结果,表明所建预测模型的预测精度较高且离散性小,其R2为91%,预测值与试验值比值的均值为0.99,标准差为0.27.此外,采用SHAP(SHapley Additive exPlanations)可解释性方法对机器学习模型预测结果进行全局解释和局部解释,特征重要性排序由重要到一般依次为:加载板宽度、截面高度、混凝土抗压强度,表明所建模型及可解释性方法是符合机理的.Reinforced Concrete(RC)is susceptible to brittle shear failure,and diagonal cracking occurs and develops throughout the process,and cracking shear is one of the key indicators.In this paper,a database of 276 shear tests with cracking load information of RC deep beams was established,and the machine learning XGBoost ensemble algorithm was used to predict their cracking strength,and five statistical indicators were used to evaluate the prediction performance of the machine learning model.The prediction results of the proposed machine learning model were compared with five semi-empirical and semi-theoretical formulations according to the case of the deep beam without web reinforcement and with web reinforcement,indicating that the prediction accuracy of the proposed prediction model is high and the dispersion is small.The R2 is 91%,the mean value of the ratio of predicted to tested values is 0.99,and the standard deviation is 0.27.In addition,SHAP(SHapley Additive exPlanations)interpretable method was used to globally and locally interpret the prediction results of the machine learning model,and the ranking of feature importance from important to general is:loading slab width,section height,and concrete compressive strength.The results show that the proposed model and interpretable method are consistent with the mechanism.

关 键 词：钢筋混凝土 深受弯构件 开裂剪力 机器学习方法 SHAP 

分 类 号：TU375.1[建筑科学—结构工程]