预制装配式型钢混凝土梁抗剪承载力的智能模型研究

The Shear Capacity Intelligent Model for Prefabricated Steel Reinforced Concrete Beams

通过建立计算预制装配式型钢混凝土(PSRC)梁抗剪承载力的智能模型,在一定程度上提高了计算精度与适用性。基于BP人工神经网络算法,通过对影响PSRC梁抗剪承载力的相关参数进行梳理,选取14个主要影响参数作为输入层,以试算法确定隐含层节点数为5,初步构建了3层结构人工神经网络系统;以收集的76组试验数据作为学习样本,对构建的神经网络系统进行训练,建立了对PSRC梁及SRC梁抗剪承载力计算的N14-5-1智能模型。使用智能模型对6个PSRC梁构件及6个SRC梁构件进行抗剪承载力计算,并通过与规范公式计算结果、试验结果的对比分析,证明了智能模型具有良好的计算精度及较好的泛化能力,具有一定的工程参考意义。运用Garson算法对输入参数进行敏感性分析,结果表明箍筋间距、型钢屈服强度、箍筋屈服强度、型钢腹板含钢率对抗剪承载力影响较大。随着研究试验的开展,在收集更多具有代表性的试验数据以扩充学习样本后,可对智能模型进一步优化。

By establishing an intelligent model for calculating the shear capacity of prefabricated steel reinforced concrete(PSRC)beams,the calculation accuracy and applicability are improved.After combing through the parameters that affect the shear bearing capacity of PSRC beams,14 main influencing parameters were selected.Based on the BP artificial neural network algorithm,with 14 main influencing parameters as the input layer,a three-layer artificial neural network system was initially constructed,using a trial-and-error method to determine that the number of nodes in the hidden layer is 5.With 76 groups of experimental data collected as learning samples,the neural network system was trained,and the N14-5-1 intelligent model for calculating the shear capacity of PSRC beams and SRC beams was established.The shear capacity of six PSRC beams and six SRC beams were calculated by using the intelligent model.The comparison between the calculation results of the standard formula and the test results proves that the intelligent model has a good calculation accuracy and good generalization ability,which has a certain engineering reference significance.The Garson algorithm is used to analyze the sensitivity of input parameters.The results show that the stirrup distance,steel yield strength,stirrup yield strength and steel ratio of section web have more influence on the shear capacity.With the development of the research and experiment,the intelligent model can be further optimized after collecting more representative experimental data to expand the learning sample.