面向通信设备信号异常识别的深度学习算法  

Deep Learning Algorithm for Signal Anomaly Recognition in Communication Equipment

作　　者：王锦毅[1,3] 茆政吉 WANG Jin-yi;MAO Zheng-ji(Concord University College,Fujian Normal University,Fuzhou Fujian 350117,China;Department of Robotics,Putian University,Putian Fujian 351100,China;School of Physics and Information Engineering,Fuzhou University,Fuzhou Fujian 350116,China)

机构地区：[1]福建师范大学协和学院,福建福州350117 [2]莆田学院机器人系,福建莆田351100 [3]福州大学物理与信息工程学院,福建福州350116

出　　处：《计算机仿真》2025年第1期215-218,228,共5页Computer Simulation

基　　金：福建省自然科学基金资助项目(2012J01295)。

摘　　要：通信设备信号可能受到多种干扰,例如电磁干扰、电源噪声等,会对信号进行扭曲和干扰,影响异常识别的准确性。现提出面向通信设备信号异常识别的深度学习算法。采用基于相似性矩阵的信号盲源分离方法将通信设备原始信号中的有用信号从背景噪声中分离出来,完成信号的去噪处理;通过自适应噪声补偿聚合经验模态分解算法分解通信设备信号,结合综合评价指标选取有效IMF分量作为信号特征;将信号特征输入卷积神经网络中,通过深度学习信号特征实现通信设备信号异常识别。通过测试发现,所提算法可在噪声背景下有效分离出有用信号,识别精度高、识别效率高。Communication equipment signals may be subject to various interferences,such as electromagnetic interference,power noise,etc.,which can distort and interfere with the signal,thus affecting the accuracy of anomaly recognition.In this paper,a deep learning algorithm for recognizing abnormal signal of communication equipment was proposed.Firstly,a blind source separation method based on similarity matrix was used to separate useful signal in the original signals of communication equipment from background noise,thus completing the signal denoising.Moreover,the signal of communication equipment was decomposed by combining adaptive noise compensation with empirical mode decomposition algorithm.Based on comprehensive evaluation indicators,effective IMF components were selected as signal features.Furthermore,signal features were input into convolutional neural network.Finally,the deep learning of signal features was used to achieve recognition for abnormal signal of communication equipment.Through the tests,the proposed algorithm can effectively separate useful signals from noisy background,with high recognition accuracy and efficiency.

关 键 词：通信设备信号 信号盲源分离 经验模态分解 卷积神经网络 深度学习 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]