基于神经网络预测与小波变换的结构非线性振动参数识别

Parametric identification of structural nonlinear vibration based on neural network prediction and wavelet transformation

首先介绍利用复Morlet小波变换进行结构非线性振动模型参数识别的原理,进而分析了因小波变换过程中的边端效应以及在采样点较少情况下复Morlet小波变换对非线性模型参数识别准确性的影响。然后提出了利用BP神经网络对非线性模型参数识别的信号进行预测延拓,并基于预测后的信号进行参数识别。最后通过对两种非线性振动模型进行数值仿真,验证了该方法能很好的提高非线性模型参数识别的准确性,并且具有一定的抗噪能力。

Here, the principle of parametric identification of structural nonlinear vibration model based on the complex Morlet wavelet transformation was introduced firstly. The influence of the complex Morlet wavelet transformation on the identification accuracy of nonlinear model parameters was analyzed under different cases of edge-effect of wavelet transformation and less sampling points. Then, a BP neural network was used for the prediction extension of nonlinear vibration signals and a novel parametric identification method was proposed based on the prediction results. Finally, through the numerical simulation of two nonlinear vibration models, the method was proved to be effective in ability identification of nonlinear model parameters and have a certain anti-noise ability.