摘要
在当前信号的盲源分离中,传统“两步法”易陷入局部最优解,并且其准确率会随采集信号数的增加或稀疏性的降低而大幅下降。针对上述问题,提出一种结合K均值-粒子群(K-PSO)和分段正交匹配追踪(StOMP)的稀疏分量分析方法。对采集信号执行K均值聚类算法,将产生的结果反馈至PSO聚类中估计混合矩阵。在获得混合矩阵后,将其源信号矩阵转化成列数为1的向量,再通过分段正交匹配追踪算法重构源信号。将实测的往复压缩机正常信号和3种单一故障信号混合成2种复合故障信号,并对复合故障信号进行试验验证。结果表明:在计算时间方面,相较模糊C均值聚类(0.335 s)和K均值聚类(0.299 s),尽管K-PSO聚类方法牺牲了一部分效率(1.561 s),但在总体角度偏差和归一化均方根误差方面表现更优,具有更好的估计精度;相较最短路径法(0.123 s),StOMP算法同样牺牲效率(2.031 s),却获得更佳的相关系数和均方根误差,表现更好的分离重构能力。这说明,该方法在盲源分离中具有可行性和实际应用价值。
In the current blind source separation of signals,the traditional"two-step method"is prone to getting stuck in local optima,and the accuracy will significantly decrease with the increase of the number of collected signals or the decrease of sparsity.In response to the above issues,a sparse component analysis method combining K-means particle swarm optimization(K-PSO)and stagewise orthogonal matching pursuit(StOMP)was proposed.The K-means clustering algorithm was performed on the collected signals,and the results were fed back to the PSO cluster to estimate the mixing matrix.After obtaining the mixed matrix,the source signal matrix was transformed into a vector with column number of 1,and then the source signal was reconstructed using the stagewise orthogonal matching pursuit.The measured normal signal of the reciprocating compressor and three single fault signals were mixed into two composite fault signals,and the compound fault signals were verified by experiments.The results show that in terms of calculation time,compared to fuzzy C-means clustering(0.335 s)and K-means clustering(0.299 s),K-PSO clustering method sacrifices some efficiency(1.561 s),but has better performance in the overall angle deviation and normalized root mean square error,and has better estimation accuracy.Compared to the shortest path method(0.123 s),the StOMP algorithm also sacrifices efficiency(2.031 s),but obtains better correlation coefficients and root mean square error,and shows better separation and reconstruction ability.This indicates that the proposed method has feasible and practical application value in blind source separation.
作者
王金东
马智超
赵海洋
李彦阳
张宇
WANG Jindong;MA Zhichao;ZHAO Haiyang;LI Yanyang;ZHANG Yu(School of Mechanical Science and Engineering,Northeast Petroleum University,Daqing Heilongjiang 163000,China)
出处
《机床与液压》
北大核心
2025年第3期228-234,共7页
Machine Tool & Hydraulics
基金
黑龙江省自然科学基金项目(LH2021E021)
黑龙江省重点研发计划项目(JD2023SJ23)。
关键词
往复压缩机
欠定盲源分离
K均值聚类
粒子群算法
分段正交匹配追踪
reciprocating compressor
underdetermined blind source separation
K-means clustering
particle swarm optimization algorithm
stagewise orthogonal matching pursuit
