基于静力数据下RBF神经网络梁式模型修正

Correction of beam structure model based on RBF neural network based on static data

基于有限元软件ABAQUS建立的梁式结构模型,并在实验室浇筑对应的实体简支梁模型。利用MATLAB神经网络工具箱构建RBF神经网络,提出一种基于静力测试数据下的径向基神经网络模型修正方法。将均匀分布的7个测点挠度作为输入向量,以对应的弹性模量、惯性矩等设计参数修正值作为输出向量,利用均匀样方设计构建神经网络样本设计表,最后利用神经网络良好的泛化能力逼近两者之间的非线性映射关系来预测修正值并与BP神经网络的修正结果进行对比。实验数据表明了修正算法的有效性。

The beam structure model was established based on the finite element software ABAQUS,and the corresponding solid simply supported beam model was poured in the laboratory.The RBF neural network was constructed by using MATLAB neural network toolbox,and a radial basis neural network model modification method based on static test data was proposed.7point delfection will be evenly distributed as input vector,with the corresponding design parameters such as elastic modulus and moment of inertia of the revised as the output vector,using the uniform quadrat design table sample build neural network design,finally using neural network generalization ability and good approximation to the nonlinear mapping relationship between forecast revisions and comparing with the results of bp neural network correction.Experimental data show the effectiveness of the modified algorithm.