基于EEMD-SOBI的水电机组多源信息分离处理

Multi-source information separation of a hydroelectric generating set based on EEMD-SOBI

水电机组振动观测信号包括相互耦合的水-机-电振源及各类噪声成分,本文提出采用集合经验模态分解-二阶盲辨识(ensemble empirical mode decomposition-second order blind source separation, EEMD-SOBI)的方法对多源观测信号进行识别。对观测信号进行解相关等初步处理后,白化计算各信号二阶统计量,计算观测信号协方差对角矩阵,最终计算振源的最优估计,对振源成分进行识别。仿真计算和模拟计算的结果均表明,仅利用观测信号均可分离出源信息且对噪声不敏感,基本能够识别出源信息,针对某电站实测单信号和多信号分析时,可有效识别出信号源成分,为水电机组的振源识别提供支撑。

According to the Complex vibration element of a hydropower unit, ensemble empirical mode decomposition-second order blind source separation(EEMD-SOBI) method, namely, EEMD-SOBI was proposed to identify multisource vibration signals. The vibration source components were recognized by the primary decorrelation of the observation signals, the whitening calculation of the second-order statistics of each signal, then the compute of the diagonalization matrix, and the optimal estimation of the vibration source.Results show that the observation signals are used to decompose and identify basically the source information with insensitivity to noise. The existing problems for directly application of SOBI are the frequency diffusion, the non-full rank of coefficient matrix, the decorrelation preprocessing of observation signals and so on. The above problems are well resolved recently, so SOBI can be exploited for vibration source analysis of hydropower unit vibration testing.