基于优化VMD融合MFCC特征的海缆裸露状态识别

Identification of Submarine Cable Exposed State Based on Optimized VMD Fusion MFCC Features

针对分布式光纤传感信号噪声强、识别难的问题,本文提出一种基于优化变分模态分解(Variational Mode Decomposition,VMD)融合梅尔频率倒谱系数(Mel Frequency Cepstral Coefficient,MFCC)特征的海缆裸露状态识别方法,用于识别海上风机海底电缆接入端浅埋和裸露两种状态。首先,利用参数优化的VMD对光纤振动信号进行分解,并利用相关系数法筛选本征模态分量(Intrinsic Mode Function,IMF);其次,融合梅尔频率倒谱系数、原始振动信号和所选IMF的时域和频域特征,以及IMF的能量和熵特征构建高维特征集,利用补偿距离评估技术(Compensation Distance Evaluation Technique,CDET)进行降维;最后,设计长短时记忆网络(Long Short Term Memory Network,LSTM)结构,将训练集输入网络进行训练,测试集验证网络的有效性,实现海缆裸露状态识别。通过现场采集的海缆振动数据进行验证,测试准确率达到100%,结果表明该方法能够准确识别和预测海缆裸露状态。

Aiming at the problem of strong noise and difficult identification of distributed optical fiber sensing signals,a method for identifying the exposed state of submarine cables based on optimized VMD and MFCC features is proposed,which is used to identify the shallow buried and exposed states of the submarine cable access end of offshore wind turbines.Firstly,the fiber optic vibration signal is decomposed using parameter optimized VMD,and the IMF is selected using the correlation coefficient method.Secondly,a high-dimensional feature set is constructed by integrating the MFCC,the original vibration signal,the time-domain and frequency-domain features of the selected IMF,as well as the energy and entropy features of the IMF.CDET is used for dimensionality reduction.Finally,design a LSTM structure,input the training set into the network for training,and validate the effectiveness of the network with the test set to achieve recognition of the exposed state of submarine cables.Through on-site collection of submarine cable vibration data for verification,the testing accuracy reached 100%,and the results showed that this method can accurately identify and predict the exposed state of submarine cables.