基于遗忘因子和LMBP神经网络的混合试验在线模型更新方法

On-line model updating method for hybrid testings based on the forgetting factor and LMBP neural network

目前将神经网络应用于混合试验的在线模型更新是一个重要的研究方向,如何提高神经网络在线模型更新算法的自适应性、稳定性和抗噪声能力是一个关键问题,提出了一种基于遗忘因子和LMBP神经网络的混合试验在线模型更新方法,即每时步利用试验子结构的历史试验数据形成带有遗忘因子的动态窗口样本,并采用增量训练方式训练LMBP神经网络,同步预测具有相同本构模型的数值子结构的恢复力。对一个两自由度非线性结构进行模型更新混合试验数值模拟,数值子结构恢复力预测值的RMSD最终为0.0230。结果表明,基于遗忘因子和LMBP神经网络的混合试验在线模型更新方法具有良好的自适应性、稳定性和抗噪声能力。

Recently the application of neural networks to the online model updating of hybrid testings is an important research direction.How to improve the adaptability,stability and anti-noise ability of online model updating algorithm of neural network is a key problem.An on-line model updating method for hybrid testings based on the forgetting factor and LMBP neural network was proposed,namely in each time step the historical experimental data of the test substructure were used to form a dynamic window sample with a forgetting factor.Then the LMBP neural network was trained with the sample set by the incremental training method,and the restoring force of the numerical element with the same constitutive model was predicted synchronously.The model updating hybrid testing on a 2-DOF nonlinear system was simulated and the RMSD of the predicted restoring force of numerical substructure was found to be 0.0230 finally.The results show that the online model updating method of hybrid testings based on the forgetting factor and LMBP neural network has good adaptability,stability and anti-noise ability.