基于IEEMD与LS-SVM组合的短期风电功率多步预测方法

针对组合预测方法中经验模态分解(EMD)部分存在处理非线性和非稳态信号的不足,提出了一种改进的集总经验模态分解(IEEMD)与最小二乘支持向量机(LS-SVM)模型相结合的短时风电功率预测方法。该方法首先通过对加噪辅助分解方法噪声准则的研究,推导出加噪方式采用正负成对形式可以有效消除分量中的残余噪声,且确定加噪幅值和分解次数采取固定值:0. 014 SD和2次。然后将原始数据通过IEEMD方法分解成一系列固有模态函数,运用游程判定法进行筛选重构成高中低频三种频段,并对不同频段的分量建立LS-SVM多步预测模型,最后将预测值自适应叠加作为最终的预测结果。通过仿真实验和实测风电功率实验验证了所提方法在预测精度上具有一定优势,为短时预测方法提供了一种新思路。     

Hidden defect recognition based on the improved ensemble empirical decomposition method and pulsed eddy current testing

This article presents a new approach for recognizing hidden defects in conductive structures based on the improved ensemble empirical decomposition (IEEMD) and the pulsed eddy current (PEC) testing technique. First, the ensemble empirical mode decomposition (EEMD) method is improved in terms of envelope fitting, end effects and false components. Then, feature vector composed of the principal marginal spectrum peaks are extracted by the help of Hilbert weighted frequencies and characteristic frequencies. It has been verified by simulation signals that the IEEMD method can obtain more accurate intrinsic mode function components (IMFs). The proposed feature extraction method may recognize various hidden defect in the rail specimen with satisfying accuracy and robustness to such disturbances as noise and liftoff.