基于EEMD和SVR的多自由度结构状态趋势预测

Trend prediction of multiple-degree of freedom structure’s state based on SVR

由于工程结构的复杂性和引起结构损伤原因的不确定性,结构早期微弱和潜在的损伤难以识别和预测。为此提出了基于聚类经验模式分解（EEMD）和支持向量机回归（SVR）的结构健康状态趋势预测方法。首先对多自由度结构渐进损伤的加速度振动信号进行聚类经验模式分解（EEMD）;再进行希尔伯特变换（HT）计算瞬时频率;然后用回归支持向量机对反映结构健康状态的瞬时频率进行趋势预测。详细分析了各种参数对回归和预测精度的影响,提出了这些参数的选用方法和一般原则。研究表明：该方法具有训练样本少的特点;在采用二阶多项式核函数、回归步长m=3~5、误差惩罚因子C=100、敏感因子ε=0.01时,可以准确地和高精度地预测结构状态趋势,预测精度达到0.24781%。

Due to the complexity of civil structure and the uncertainty of factors caused structural damage,it is difficult to identify and predict the early faint and potential damage.So a trend prediction model of support vector machine(SVM) based on ensemble empirical mode decomposition(EEMD) is proposed.Firstly,the response signals of acceleration of a multi-degree of freedom structure model are processed by using EEMD,the intrinsic mode function(IMF) which contains structural damage information are selected;secondly,the selected IMF is transformed by using Hilbert transform(HT) and instantaneous frequency(IF) are calculated;thirdly,the trend prediction based on SVR of IF is realized;finally,the effects of various parameters on the regression and prediction accuracy are analyzed in detail,the choice methods and general rules for selecting these parameters are proposed.The prediction of structural engineering simulation data shows that the method with less training samples can predict the trends of structure conditions accurately and precisely,the forecasting accuracy is within 0.24781%,when the kernel function is the second order polynomial kernel function,the length of return step m is 3~5,the error of punish factor C is 100,sensitive factor ε is 0.01.