基于TSVD-SCN的光纤入侵信号识别算法研究

A recognition algorithm of optical fiber intrusion signals based on TSVD-SCN neural network

采用随机配置网络(SCN,Stochasti cconfiguration network)对光纤振动信号进行识别,常由于光纤预警系统的背景噪声问题使得网络的隐含层输出接近奇异,直接影响了SCN对光纤数据的识别准确率。因此本文提出了一种基于截断奇异值分解(Truncated singular value decomposition,TSVD)的SCN方法(TSVD-SCN)对光纤入侵信号进行识别。TSVD-SCN通过对网络的隐含层输出进行SVD分解并设置阈值去除其中较小的奇异值,以减少隐含层输出矩阵的条件数,提升网络识别率。本文利用占空比,平均幅差函数,FFT求能量占比的方法进行特征提取,采用基于TSVD-SCN算法对不同入侵振动特征矢量进行分类识别。实验证明,本文所提算法模型精度比SCN的模型精度更高,可以准确识别光纤入侵信号类型,对SCN网络在实际应用中对分类精度的提高有着重要意义。

Because of the background noise problem of optical fiber early warning system,the hidden layer output of the network is close to singularity,and the recognition accuracy is low when Stochastic configuration Network(SCN) is used to identify the optical fiber vibration signal.Therefore,a SCN method(TSVD-SCN) based on the Truncated singular value decomposition(TSVD-SCN) is proposed in this paper to identify the optical fiber intrusion signals.TSVD-SCN performs SVD decomposition on the hidden layer output of the network and sets thresholds to remove the smaller singular value,reducing the condition number of the output matrix of the hidden layer,and improving the network recognition rate.This paper utilizes the function of duty cycle,average magnitude difference function and the frequency domain energy ratio are used to extract the different intrusion features of multiplex signal,respectively.The classification of the feature vectors for different intrusion vibrations is realized by using the TSVD-SCN algorithm.Experimental results show that the proposed algorithm has higher accuracy than that of SCN model,and can accurately identify the type of optical fiber intrusion signal.It is significant to improve the classification accuracy of SCN network in frequency domain.