基于深度学习的框架结构首层位移角预测技术

PREDICTION OF DISPLACEMENT ANGLE OF FIRST FLOOR OF FRAME STRUCTURE BASED ON DEPTH LEARNING TECHNOLOGY

基于深度学习的方法,首先拟合现浇框架节点弯矩转角曲线,得到同等条件下装配式框架的节点刚度,然后对一榀现浇及装配式框架进行数值模拟,得到2种框架结构在同一地震荷载作用下的首层层间位移角,用神经网络拟合同一条件下现浇及装配式框架首层层间位移角的二维曲面关系。通过有限元模拟验证预测结果精度较高,该网络可用于装配式框架结构的首层层间位移角预测,减少计算成本,节约时间,适用于工程实践。

Based on the depth learning method,this paper first fits the moment angle curve of cast-in-place frame node,The node stiffness of the fabricated frame under the same conditions is obtained.Then a cast-in-place and fabricated frame is numerically simulated to obtain the first floor displacement angle of the two frame structures under the same seismic load.The neural network is used to simulate the two-dimensional curved surface relationship of the first floor displacement angle of the cast-in-place and fabricated frames under the same conditions.The precision of prediction results is high through finite element simulation.The network can be used to predict the first floor displacement angle of fabricated frame structure,reduce calculation cost,save time,and is more suitable for engineering practice.