基于SSA-RBFNN的钢管混凝土界面粘结强度研究  

Study on interfacial bond strength of concrete-filled steel tube based on SSA-RBFNN

作　　者：刘文博 杨喜娟 王力[1] 李子奇[1] LIU Wenbo;YANG Xijuan;WANG Li;LI Ziqi(School of Civil Engineering,Lanzhou Jiaotong University,Lanzhou Gansu 730070,China;School of Architectural Engineering,Baiyin Vocational College of Mining and Metallurgy,Baiyin Gansu 730900,China)

机构地区：[1]兰州交通大学土木工程学院,甘肃兰州730070 [2]白银矿冶职业技术学院建筑工程学院,甘肃白银730900

出　　处：《中国安全生产科学技术》2025年第3期148-155,共8页Journal of Safety Science and Technology

基　　金：甘肃省自然科学基金青年项目(25JRRA207);兰州铁路局集团公司科技研究开发计划项目(LZJKY2024041-1);山西交通控股集团科技创新项目(23-JKKJ-6)。

摘　　要：为了改善传统径向基神经网络(RBFNN)存在对样本数据依赖性强、参数选择复杂、收敛速度慢等缺陷。将麻雀搜索算法(SSA)应用于RBFNN预测模型,提出基于SSA-RBFNN的CFST界面粘结强度预测模型,收集319组数据建立数据库,选取8种影响因素作为输入层参数和界面粘结强度作为输出层参数,分别建立RBFNN和SSA-RBFNN模型。通过平均绝对百分比误差(MAPE)和决定系数(R 2)等指标,将2种机器学习模型与6种现有公式进行比较,评估它们在预测精度和稳定性方面的表现。研究结果表明:2种机器学习模型比公式精度更高。其中,SSA-RBFNN模型有更好的预测性能,更有助于高效预测CFST的界面粘结强度。研究结果可为CFST结构工程设计提供相应的预测方法和技术支持,可以帮助工程师在设计和施工过程中更好地评估结构的承载能力和安全性。To address the shortcomings of traditional radial basis function neural networks(RBFNN),such as strong dependence on sample data,complex parameter selection,and slow convergence speed,the sparrow search algorithm(SSA)was applied to RBFNN prediction model,and a prediction model for the interfacial bond strength of concrete-filled steel tube(CFST)based on SSA-RBFNN was proposed.A database consisting of 319 sets of data was established,then eight types of influencing factors were selected as the input layer parameters,and the interfacial bond strength was selected as the output layer parameter.Both the RBFNN and SSA-RBFNN models were developed accordingly.By using metrics such as mean absolute percentage error(MAPE)and coefficient of determination(R 2),two machine learning models are compared with six existing formulas to evaluate their performance in terms of prediction accuracy and stability.The results show that both the machine learning models outperform the formulas in terms of accuracy.The SSA-RBFNN model has the better prediction performance,and it is more helpful to predict the interfacial bond strength of CFST efficiently.The research results can provide corresponding prediction methods and technical support for CFST structural engineering design,and can help engineers better evaluate the bearing capacity and safety of the structure during design and construction.

关 键 词：RBF神经网络 麻雀搜索算法 钢管混凝土 界面粘结强度 机器学习模型 

分 类 号：X947[环境科学与工程—安全科学]