(1 河北建筑工程学院土木工程学院, 张家口 075031; 2 河北省土木工程诊断,改造与抗灾重点实验室, 张家口 075031; 3 河北工业大学土木与交通学院,天津 300401; 4 国网冀北电力有限公司张家口供电公司, 张家口 075001)

摘要: 筛选文献中报道的 143 组试验数据,采用误差反向传播人工神经网络(BP-ANN)建立和训练了 1 个 2 层BP-ANN 模型,对梁柱平齐端板连接节点的极限抗弯承载力进行了预测. 该模型利用 20 个组件特征参数作为输入,以极限抗弯承载力作为输出. 通过与传统机器学习算法预测结果对比,验证了方法和模型的有效性,并依据模型推导出一个实用简化的极限抗弯承载力数学表达式. 统计分析结果显示:经过训练的 BP-ANN 模型,在测试集上的平均绝对百分比误差(MAPE)为 5. 28%,均方误差(MSE)为 5. 79×10-4. 另外,对特征参数进行敏感性分析,得到了组件特征对节点极限抗弯承载力的影响程度. 研究结果表明:采用 BP-ANN 模型能够综合考虑组件特征对节点极限抗弯承载力的影响,预测结果较为准确;该模型为梁柱连接性能评估和改进提供了智能化的解决方案,可作为数值模拟和结构试验研究的有力补充.

关键词: 人工神经网络; 梁柱节点; 平齐端板连接; 极限抗弯承载力

中图分类号:TU391 文献标志码:A 文章编号:1002-848X(2023)05-0119-08

DOI:10. 19701 / j. jzjg. LS220111

Research on prediction of ultimate moment capacity of beam-to-column flush end-plate connection joint via BP-ANN model

LIU Zhongyang1,2,3, LI Bingyang1, HUANG Yimiao3, WANG Hao3, CHEN Wei4, WENG Weisu1,2

(1 School of Civil Engineering, Hebei University of Architecture, Zhangjiakou 075031, China; 2 Hebei KeyLaboratory for Diagnosis, Reconstruction and Anti-disaster of Civil Engineering, Zhangjiakou 075031, China;3 School of Civil and Transportation Engineering, Hebei University of Technology, Tianjin 300401, China;4 State Grid Zhangjiakou Power Supply Company, Zhangjiakou 075001, China)

Abstract: Based on 143 sets of experimental data reported in the literatures, a two-layer back-propagation artificial neuralnetwork (BP-ANN) model was established and trained to predict the ultimate moment capacity of beam-to-column flushend-plate connection joint. The model uses 20 component features as the input and the ultimate moment capacity as theoutput. Compared with the prediction results of traditional machine learning algorithms, the effectiveness of the method andmodel were verified, and a simplified explicit expression of ultimate moment capacity was derived. The statistical analysisresults show that the trained BP-ANN model has a mean absolute percentage error (MAPE) of 5. 28 % and a mean squareerror (MSE) of 5. 79×10-4on the test set. In addition, the influence of component features on the ultimate moment capacityof joint was obtained by sensitivity analysis. The study results show that the BP-ANN model can comprehensively consider theinfluence of component features on ultimate moment capacity of joint, and the prediction results are more accurate. The modelprovides an intelligent solution for the performance evaluation and improvement of beam-to-column connections, and it can beused as a powerful supplement to the numerical simulation and structural test research of such connections.

Keywords:artificial neural network; beam-to-column connection joint; flush end-plate connection; ultimate moment capacity

 ∗河北省省属高等学校基本科研业务费研究项目(2021XSTD08).

第一作者:刘仲洋,博士研究生,副教授,一级注册结构工程师,主要从事装配式建筑钢结构的研究,Email:lzy1502@ hebiace.edu. cn.

[引用本文] 刘仲洋,李冰阳,黄轶淼,等. 采用 BP-ANN 模型的梁柱平齐端板连接节点极限抗弯承载力预测研究[J]. 建筑结构,2023,53(5):119-126. LIU Zhongyang,LI Bingyang,HUANG Yimiao,et al. Research on prediction ofultimate moment capacity of beam-to-column flush end-plate connection joint via BP-ANN model[ J]. Building Structure,2023,53(5):119-126.