基于频谱聚类分割的粒子群最优模态分解仿真研究

Simulation Study of Modal Parameter Estimation Algorithm with PSO Based on Spectrum Segmentation

在两种有力措施的基础上提出了粒子群最优模态参数识别算法。一是提出了一种性能稳定的模态参数初始值估计算法,引入模态聚类的思路来估计出各个模态参数的上下限范围。该算法把幅谱曲线看成是局部波峰的集合,按聚类分割思路来构造聚类距离函数,使用k-means算法把振动信号频谱自动聚类成多个单模态类,然后运用单模态分解算法估计出每个模态类的模态参数的上下限范围,给出粒子属性值的上下界,极大地减少粒子群算法的搜索空间,减少最优搜索时间提高搜索结果的稳定性。二是采用了混合变异粒子群算法来提高最优化搜索的效率,有效避免陷入局部最优,提高模态参数的准确性。从仿真信号的大量实验研究结果看,与经典的正交多项式拟合算法相比,该算法的噪声抵抗能力更强、更稳定。

A new modal parameter estimation algorithm using particle swarm optimization was put forward based on two powerful methods. The first powerful method is a stable initial modal parameters estimation algorithm. The estimation algorithm introduces cluster mode of spectrum into modal parameters estimation. Taking the spectrum curve as a collection of local peaks, constructing the distance function of modal class, and using k-means algorithm to segment the spectrum curve into many single modal spectrum curves, the ranges of initial modal parameters will be obtained by single modal parameter estimation algorithm. So the search space of PSO is reduced, and the results of PSO are more stable. The second powerful method is using the hybrid mutation PSO algorithm. The hybrid mutation PSO algorithm can search results more quickly and more stably. By the amount of experimental results of simulation signals, compared to the classic orthogonal polynomials fitting algorithm, the algorithm is stronger noise resistance and more stable.