在线神经网络算法研究及其在混合试验中的应用

发布时间：2018-03-21 05:45

本文选题：混合试验　切入点：神经网络　出处：《黑龙江科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：混合试验是最近几年新型的将数值模拟与真实加载试验相结合的结构抗震试验方法,得到了相关研究者们的广泛关注。混合试验中,一般将结构中关键部位作为试验子结构进行试验,其他部位作为数值子结构进行数值模拟。为了模拟具有强非线性的数值子结构,本文提出一种采用在线神经网络算法,并进行了防屈曲支撑构件试验的恢复力模型在线预测,最后提出了基于在线神经网络算法的混合试验方法。具体研究内容如下:1、提出一种在线BP神经网络算法,在线预测非线性结构恢复力。针对两个自由度非线性结构,进行了Bouc-Wen模型的恢复力预测,验证了在线神经网络算法的有效性。分析了样本数量以及目标误差对预测精度和计算耗时的影响。研究表明:与传统离线学习方式算法相比,在线BP神经网络提高了算法计算效率和预测精度。2、基于递推形式的在线Elman神经网络的防屈曲支撑构件试验,研究表明:基于递推形式的Elman神经网络算法具有较好的实时性和动态性,能够有效提高恢复力预测精度。3、基于在线神经网络算法的混合试验方法,通过两层非线性模拟结构的混合试验数值预测得出该方法的可行性。试验结果表明,与在线BP网络算法相比,在线Elman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Elman算法的结构混合试验方法可以提高混合试验结果精度。
[Abstract]:Hybrid test is a new seismic test method combining numerical simulation and real load test in recent years, which has been paid much attention by researchers concerned. In order to simulate the numerical substructure with strong nonlinearity, an on-line neural network algorithm is proposed to simulate the key parts of the structure as experimental substructures and other parts as numerical substructures. Finally, a hybrid test method based on on-line neural network algorithm is proposed. The specific research contents are as follows: 1, an online BP neural network algorithm is proposed. The Bouc-Wen model is used to predict the restoring force of nonlinear structures with two degrees of freedom. The effect of sample size and target error on prediction accuracy and computation time is analyzed. The results show that the proposed algorithm is more effective than the traditional off-line learning algorithm. The calculation efficiency and prediction accuracy of the algorithm are improved by on-line BP neural network. The buckling bracing test of on-line Elman neural network based on recursive form is carried out. The results show that the Elman neural network algorithm based on recursive form has better real-time and dynamic characteristics, and can effectively improve the prediction accuracy of recovery force. The hybrid test method based on on-line neural network algorithm is proposed. The feasibility of this method is obtained by numerical prediction of mixed experiments of two-layer nonlinear simulation structure. The experimental results show that the on-line Elman algorithm is more adaptive than the on-line BP network algorithm. It can effectively improve the prediction accuracy and computational efficiency of the restoring force, and the hybrid structure test method based on online Elman algorithm can improve the accuracy of the mixed test results.
【学位授予单位】：黑龙江科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TU352.11

【参考文献】