针对大坝内部垂直位移数据序列不仅具有周期性、平稳性,且在频域上存在高、低频,显著的多尺度等特性,本实验利用多尺度小波分析的原理与方法对数据序列进行分解,对低频序列和高频序列分别建立AR模型和BP神经网络模型并进行预测,最后叠加各个序列的预测值,得到数据序列的预测结果。该方法适用于大坝垂直位移的预测,结果与自回归模型及单BP神经网络模型相比,该模型具有更高预测精度。%According to the internal vertical displacement data a sequence is periodic and smooth with multiscale features significantly . This experiment uses the principle and method of wavelet multi‐scale analysis to decompose the vertical displacement data and reconstruct the low frequency and high frequency sequence AR model and BP neural network models .By fitting models for forecasting ,the predictions of each sequence are obtained .The method is applied to the predict the dam vertical displacement ,and the results are contrasted with single autoregressive model and the BP neural network model ,which proves to be with higher predictive precision .
展开▼