基于LMD与样本熵的多尺度大坝变形预测

摘要

针对大坝变形数据的多尺度特征，将局域均值分解、样本熵及高斯过程算法应用于大坝变形预测中，提出了多尺度大坝变形预测新模型。首先利用局域均值分解算法对变形数据进行多尺度分析，挖掘变形数据隐含的信息，随后根据各变形分量特征，构建基于高斯过程的多尺度大坝变形预测模型，并利用样本熵对模型进行简化。通过实例分析，证实该大坝变形预测新方法精度高于BP网络和最小二乘支持向量机模型。%Aiming at the multi-scale characteristics of dam deformation data, a new model for multi-scale deformation fore-casting was proposed based on the local mean decomposition, sample entropy and Gaussian process algorithm. Firstly, the de-formation data are analyzed by the local mean decomposition algorithm to find out the implicit information;then the multi-scale prediction model for dam deformation is built based on Gaussian process according to the characteristics of each deformation com-ponent, and the model is simplified by sample entropy. The experimental results prove that the new prediction method is better than BP and SVM model in accuracy.