• 1
  • 原文摘要
  • 混合试验是一种将数值模拟与物理试验相结合的新兴结构抗震试验方法,得到了相关研究者们的广泛关注.如何模拟具有强非线性的数值子结构仍是混合试验亟待解决的问题.在传统的离线神经网络基础上提出一种在线学习的神经网络算法,并应用于混合试验中来在线预测数值子结构恢复力.在线学习算法仅利用当前步的系统输入和观测样本,采用递推形式更新每一步的权值和阈值.针对两个自由度非线性结构,分别进行了基于在线学习和离线学习神经网络的混合试验数值仿真.研究表明:与离线学习神经网络算法相比,在线学习神经网络算法具有更好的自适应性,能够有效提高恢复力预测精度和计算效率;基于在线学习神经网络算法的结构混合试验方法可以提高混合试验结果精度. Hybrid testing is an advanced structural seismic experimental method that combines the numerical simulation and physical testing and is increasingly being recognized by researchers.One of the challenging issues is how to model the numerical substructure with strong nonlinearity.An online learning neural network (NN) algorithm was proposed based on the conventional offline NN algorithm and applied in the hybrid testing to online predict the restoring force for the numerical structure.Weights and thresholds of the novel algorithm can be updated using a recursive form only based on the current step systematic inputs and observations.Numerical simulations for the hybrid testing based the online and the offline NN algorithms were conducted on a two degrees of freedom nonlinear system.The results show the new online NN hybrid testing method has better adaptation, computational efficiency and prediction accuracy for the detection of restoring force, compared with the offline NN hybrid testing method.In general, the online NN hybrid testing method can effectively improve the accuracy of hybrid testing results.
  • 用户摘录