Abstract: The frequency-mixing-producing mechanism is discussed and two prerequisites of its occurrence are concluded.Partial empirical model decomposition algorithm is proposed to deal with the frequency mixing problem.The algorithm locates short-time higher frequency components by probing jumps in the signal，and sets up local time windows to cover them.It adopts empirical model decomposition just within the windows to separate the short-time higher frequency components from background signal.In this way the intrinsic mode functions in the new result can reflect the temporal processes better.Partial empirical model decomposition method is simple，highly unified with the theory of empirical model decomposition.An application case shows the effective of the new method.

Key words: empirical model decomposition, intrinsic mode function, abnormal event, frequency mixing

摘要： 对经验模态分解算法中的异常事件干扰机制做了深入的探讨，指出发生频率混叠现象时必须满足的两个条件。为了避免出现频率混叠现象，提出了基于动态窗口的局部分解算法。利用信号的时间特征尺度检测出信号的突变并定位局部高频分量，在分解信号的过程中，局部分解算法并不对信号的整个时间区域进行分解，而是以定位好的局部高频分量位置为窗口，进行局部的经验模态分解，分离出高频分量。通过这种局部分解，就可以有效地消除模态间的频率混叠，得到的固有模态函数更可靠地反映了真实物理过程。和现有异常事件处理方法相比，局部经验模态分解算法在理论上和经验模态分解算法更为统一，方法更为简便。通过实例表明了局部经验模态分解算法的有效性。

关键词: 经验模态分解（EMD）方法, 固有模态函数（IMF）, 异常事件, 频率混叠