改进的气体泄漏识别算法研究  

Research on improved compressed gas leakage recognition method

作　　者：郑伟哲 韦娟[1] 郑浩南 ZHENG Weizhe;WEI Juan;ZHENG Haonan(School of Communication Engineering,Xidian University,Xi’an 710071,China)

机构地区：[1]西安电子科技大学通信工程学院,陕西西安710071

出　　处：《现代电子技术》2023年第5期35-39,共5页Modern Electronics Technique

摘　　要：为了获得理想压缩气体泄漏信号的识别准确率,进而实现高效检测的目的,提出一种基于经验模态分解、梅尔频率倒谱系数和主成分分析的泄漏超声信号特征提取方法。首先,使用经验模态分解提取泄漏信号的超声频段,通过对固有模态函数的熵值设定阈值,优化频谱混叠;其次,通过构造梅尔变换函数,设计分别针对目标频段中不同分布的梅尔滤波器组;然后,使用主成分分析代替离散余弦变换,提取改进的梅尔频率倒谱系数;最后,在实验室模拟泄漏环境,采集不同泄漏条件的泄漏信号,使用支持向量机实现识别分类,完成泄漏检测。结果表明,使用熵阈值优化的经验模态分解能够提高泄漏信号的识别准确率,改进的梅尔频率倒谱系数是一种更有效的泄漏信号特征,相比改进前识别准确率提高了7.76%。In order to obtain the ideal recognition accuracy of the compressed gas leakage signal and realize the purpose of high efficiency detection,a leakage ultrasonic signal feature extraction method based on empirical mode decomposition,Mel frequency cepstrum coefficient and principal component analysis is proposed. The empirical mode decomposition is used to extract the ultrasonic frequency band of the leakage signal,and the spectrum aliasing is optimized by setting the threshold value of the entropy value of the intrinsic mode function. Mel filter banks for the low frequency,intermediate frequency and high frequency of the target frequency band are designed respectively by constructing the Mel transform function. The principal component analysis is used to instead of discrete cosine transform to extract the improved Mel frequency cepstral coefficients.The leakage environment is simulated in the laboratory,and the leakage signals of different leakage conditions are collected. The support vector machine is used to realize the identification and classification,and complete the leakage detection. The results show that the empirical mode decomposition optimized by the entropy threshold can improve the recognition accuracy of the leaked signal,and the improved Mel frequency cepstrum coefficient is a more effective feature of the leaked signal. Its recognition accuracy is increased by 7.76% in comparison with the previous recognition accuracy.

关 键 词：气体泄漏识别 信号特征提取 阈值设定 频谱混叠优化 主成分分析 梅尔滤波器 泄漏检测 

分 类 号：TN911.6-34[电子电信—通信与信息系统] TB553[电子电信—信息与通信工程]