Abstract: Aiming at the complex multi-source excitation feature extraction problems of the seawater pumps， a feature extraction method was proposed based on the combination of SOA，VMD and ICA for the excitation sources of the seawater pumps. Based on the single-channel measurement signals， the VMD algorithm and SOA optimization algorithm were used to calculate the signal square envelope spectral kurtosis statistic as the fitness function to optimize the modal decomposition number K，the penalty coefficient α，and the eigenmodal function components（IMF). The signal alignment entropy was used as the noise test function. After choosing a reasonable threshold value， the IMF components for noise were screened then the non-noise IMF component signals were obtained. Fast-ICA algorithm was used to obtain the excitation source signal vector，thus realizing the excitation source feature signal extraction. The effectiveness of the feature extraction method was verified through the experimental and comparative analysis of the excitation source feature extraction of the real ships seawater pumps. The results show that SOA-VMD-ICA algorithm meet the requirements of extracting the excitation source features of seawater pumps accurately under single-channel measurement conditions.

Key words: feature extraction, seawater pump, independent component analysis(ICA), seagull optimization algorithm(SOA), variational modal decomposition(VMD)

摘要： 针对海水泵复杂多源激励特征提取问题，提出了一种海鸥优化算法（SOA）、变分模态分解（VMD）和独立分量分析（ICA）相结合的海水泵激励源特征提取方法。基于单通道测量信号，采用VMD算法与SOA算法选取信号平方包络谱峭度统计量作为适应度函数，寻优获取模态分解数量K、惩罚系数α及特征模态函数（IMF）分量。采用信号排列熵作为噪声检验函数，合理选取排列熵阈值，对IMF分量进行噪声筛选，获取非噪声IMF分量信号。将非噪声IMF分量与原输入信号组合，采用快速独立成分分析（Fast-ICA）算法计算得到激励源信号向量，从而实现激励源特征信号的提取。通过实船海水泵激励源特征提取试验及对比分析，验证了所提方法的有效性。研究结果表明，所提的SOA-VMD-ICA方法能满足单通道测量条件海水泵激励源特征提取准确性要求。

关键词: 特征提取, 海水泵, 独立分量分析, 海鸥优化算法, 变分模态分解