基于联合熵驱动改进麻雀搜索优化VMD的DAS信号去噪方法

Joint Entropy-Driven Improved Sparrow Search Optimized VMD for DAS Signal Denoising

摘要: 分布式光纤声波传感(Distributed Acoustic Sensing, DAS)系统采集的振动信号易受到环境噪声和系统噪声干扰，导致有效特征提取困难。针对传统变分模态分解(Variational Mode Decomposition, VMD)参数依赖经验选取以及单一评价指标难以准确表征分解质量的问题，提出一种联合熵驱动的改进麻雀搜索优化变分模态分解(EE-PE-ISSA-VMD) DAS信号去噪方法。以模态数K和惩罚因子α为优化对象，引入Tent混沌初始化、自适应发现者比例和自适应步长策略对麻雀搜索算法(SSA)进行改进，实现VMD参数自适应优化；构建包络熵(Envelope Entropy, EE)与排列熵(Permutation Entropy, PE)联合适应度函数，从信号能量分布和时间序列复杂度两个维度综合评价VMD分解质量；同时采用基于相关系数的本征模态函数(IMF)筛选重构策略，提高信号重构精度和噪声抑制能力。基于MATLAB平台，本文参考PubDAS公开数据集及相关文献中DAS振动信号的典型时域和频谱特征，构建模拟DAS振动信号，并与小波阈值去噪、传统VMD、SSA-VMD及ISSA-VMD方法进行对比实验。实验结果表明，本文方法获得的最优参数组合为K = 10、α = 646，信噪比(SNR)达到10.2629 dB，均方根误差(RMSE)降低至0.2982。与传统VMD相比，SNR提高8.11%，RMSE降低8.47%；与ISSA-VMD相比，SNR进一步提高0.4817 dB。结果表明，所提方法能够有效抑制噪声干扰，在保留有效振动特征的同时提高信号重构质量，为DAS信号降噪及智能光纤感知系统信号处理提供了一种有效技术方案。

Abstract: Distributed Acoustic Sensing (DAS) signals are susceptible to environmental and system noise interference, which makes useful vibration feature extraction difficult. To address the dependence of Variational Mode Decomposition (VMD) on empirical parameter selection and the limited ability of a single evaluation index to characterize decomposition quality, this paper proposes a joint entropy-driven Improved Sparrow Search Algorithm optimized VMD (EE-PE-ISSA-VMD) method for DAS signal denoising. The modal number K and penalty factor α are taken as optimization variables. Tent chaotic initialization, an adaptive discoverer ratio, and an adaptive step-size strategy are introduced to improve the Sparrow Search Algorithm (SSA), thereby realizing adaptive optimization of VMD parameters. A joint fitness function combining Envelope Entropy (EE) and Permutation Entropy (PE) is constructed to evaluate VMD decomposition quality from the perspectives of energy distribution and time-series complexity. In addition, a correlation coefficient-based Intrinsic Mode Function (IMF) selection and reconstruction strategy is employed to improve reconstruction accuracy and noise suppression capability. Simulation experiments were conducted on simulated DAS vibration signals constructed in MATLAB with reference to the typical time-domain and spectral characteristics of DAS vibration signals reported in the PubDAS public dataset and related literature. The experimental results show that the optimal parameter combination obtained by the proposed method is K = 10 and α = 646, achieving a Signal-to-Noise Ratio (SNR) of 10.2629 dB and a Root Mean Square Error (RMSE) of 0.2982. Compared with conventional VMD, the proposed method improves the SNR by 8.11% and reduces the RMSE by 8.47%. Compared with ISSA-VMD, the SNR is further improved by 0.4817 dB. The results demonstrate that the proposed method can effectively suppress noise interference while preserving useful vibration features and improving signal reconstruction quality, providing an effective technical solution for DAS signal denoising and intelligent fiber-optic sensing applications.