基于TensorFlow的单层球面网壳结构极限承载力预测

Prediction of Ultimate Bearing Capacity of Single-Layer Spherical Reticulated Shell Based on TensorFlow

为研究单层球面网壳结构极限承载力问题,基于TensorFlow下的BP神经网络算法,考虑非线性分析中的复杂映射关系,建立神经网络模型,对K8型单层球面网壳结构的极限承载力进行预测。在此基础上,考虑结构网格形式的不同,建立新的神经网络模型,预测K n型单层球面网壳结构的极限承载力;将预测结果与有限元和文献回归公式的计算结果进行对比分析。研究结果表明:预测的K8型单层球面网壳结构的极限承载力与有限元结果误差均值为1.666%,文献回归公式计算的结果与有限元的计算结果误差均值为3.994%;预测的Kn型单层球面网壳结构的极限承载力与有限元结果误差均值为4.774%,文献回归公式计算的结果与有限元的计算结果误差均值为5.163%。可见利用神经网络对单层网壳结构极限承载力进行预测是可行的。

In order to study the ultimate bearing capacity of the single-layer spherical reticulated shell structure,based on the BP neural network algorithm of TensorFlow,a neural network model was established to predict the ultimate bearing capacity of the K8 single-layer spherical reticulated shell structure by considering the complex mapping relationship in the nonlinear analysis.Moreover,another new neural network model was established to predict the ultimate bearing capacity of the K n-type single-layer spherical reticulated shell structure.The prediction results were compared with the calculation results of the finite element and literature regression formulas.The results showed that the error mean between the predicted ultimate bearing capacity of the K8 single-layer spherical reticulated shell structure and the finite element calculation results was 1.666%,and that between the formula calculation results and the finite element calculation results was 3.994%;the error mean between the predicted ultimate bearing capacity of the K n-type single-layer spherical reticulated shell structure and the finite element calculation was 4.774%,and that between the formula calculation results and the finite element calculation results was 5.163%.The feasibility of using neural network to predict the ultimate bearing capacity of single-layer reticulated shell structure is demonstrated.