Abstract: To improve de-noising performance of Power Quality Mixed Disturbance（PQMD） signals and solve the problem of feature detection for PQMD, a new de-noising method and mathematical framework using Adaptive Multi-resolution Singular Value Decomposition（AMSVD） are proposed. First, the distribution of singular values of Gaussian white noise decomposed by multi-resolution SVD is used to analyze de-noising theory for noisy signal, and then the rule for optimal decomposition level is given according to the difference of approximation signal and detail under different levels, through these procedures adaptive de-noising method for multi-resolution SVD can be implemented for unknown prior information signals. The results show that AMSVD is effective than other 5 filter algorithms in PQMD signals processing with different noise variance. To further verify the feasibility of AMSVD method, the characteristic of the disturbance can be extracted with Hilbert-Huang Transform（HHT）. The simulation results prove that AMSVD is feasible and robust.

Key words: Power Quality Mixed Disturbances（PQMD）, de-noising, Singular Value Decomposition（SVD）, Adaptive Multi-resolution Singular Value Decomposition（AMSVD）, feature detection

摘要： 为了提高电能质量复合扰动（PQMD）信号的去噪指标，实现扰动信号特征的准确检测，提出一种自适应多尺度SVD（Adaptive Multi-resolution Singular Value Decomposition，AMSVD）去噪新算法及数学框架。该算法首先分析了高斯白噪声奇异值分布情况及多尺度SVD消噪原理，针对不同尺度下的噪声近似与细节信号奇异值差值规律，确定出最佳消噪尺度的约束条件，由此实现噪声先验信息未知的自适应消噪方法。研究结果表明，在对不同噪声方差下的电能质量复合扰动去噪处理中，AMSVD消噪效果优于其他5种方法。为了进一步验证AMSVD算法去噪后特征量检测的准确性，采用希尔伯特黄变换（HHT）提取扰动特征信息，仿真结果表明该算法具有可行性和鲁棒性。

关键词: 电能质量复合扰动, 去噪, 奇异值分解（SVD）, 自适应多尺度奇异值分解（AMSVD）, 特征检测