中文摘要：

将Levenberg-Marquardt BP人工神经网络应用于复杂的非线性振动响应的趋势预测，避免了时序分析复杂的数据预处理、模型识别、参数估计和模型适用性检验过程。通过对样本预测效果的比较，全面考虑了网络的输入层、隐层和输出层的神经元节点数和各层之间的传递函数对预测精度的影响，引入Box-Cox变换改善了网络的收敛性并加快了网络的收敛速度，同时采用重复训练法来提高网络的稳定性和预测精度。预测实例表明，相比于传统的时间序列分析方法，这种预测方法能对振动响应的趋势进行更准确的预测。

英文摘要：

The application of Levenberg-Marquardt BP artificial neural network to the complicated nonlinear vibration response trend forecast avoids complex data pre-processing, model recognition, parameter estimation and model applicability evaluation in time series analysis. The comparison of sample forecast effects takes into full account the impacts of the number of neurons at input layer, hidden layer and output layer of the neural network and the transfer function among all the layers on the forecast accuracy. The Box-Cox transformation was introduced to improve the convergence of the neural network and quicken its convergence speed. In the meanwhile, repetitive training was conducted to enhance its stability and forecast accuracy. A forecast instance indicates that compared with the traditional time series analysis, the present forecast method can carry out more accurate forecast of vibration response trends.