改进的HHT变换在光纤振动模式识别中的应用

Application of modified Hilbert-Huang transform in fiber-optic vibration pattern recognition

针对双马赫-曾德尔(M-Z)干涉型光纤振动传感系统输出信号非线性、非平稳特点,提出基于互补总体经验模态分解(CEEMD)的希尔伯特-黄变换(HHT)模式识别算法。该算法采用CEEMD将振动信号分解成多个本征模态函数(IMF),利用各阶IMF的归一化自相关函数筛选出噪声分量进行中值滤波;然后对各阶IMF分量做Hilbert变换,基于Hilbert边际能量谱构造特征向量;最后利用概率神经网络(PNN)实现振动信号的模式识别。对四种典型光纤振动信号的实验验证表明,算法的平均正确识别率最低可达85%。

According to the nonlinear and nonstationary characteristics of output signal of dual Mach - Zenhder (M - Z) interferometric fiber.optic vibration sensing system,a modified Hilbert.Huang transform(HHT) pattern recognition algorithm based on complementary ensemble empirical mode decomposition(CEEMD)is proposed. The CEEMD is used in the algorithm to decompose the vibration signals into several intrinsic mode functions(IMFs),and the normalized autocorrelation function of each.order IMFs is adopted to screen out the noise component for median filtering. The Hilbert transform is performed for each. order IMFs component,and the feature vector is constructed on the basis of Hilbert marginal energy spectrum. The probabilistic neural network(PNN) is utilized to realize the pattern recognition of vibration signal. The experimental results of four typical fiber-optic vibration signals show that the average correct recognition rate of the algorithm can reach up to 85%.