基于EMD-AWPP和HOSA-SVM算法的分布式光纤振动入侵信号的特征提取与识别

针对传统的信号处理方法无法有效区分不同振动入侵信号，提出一种基于EMD-AWPP和HOSA-SVM算法的振动信息特征提取与识别方法，用于解决分布式光纤振动入侵检测系统的高精度信号识别问题。处理不同振动类型时，该方法首先利用基于经验模态分解的自适应小波包处理算法，不仅对信号的低频部分进行了分解，而且对高频部分即信号的细节部分也进行了更好的时频局部化处理，改善了信号特征提取精度，减少传感信号异常值的影响; 其次采用高阶谱分析中的双谱和双相干谱，精确提取包含不同振动入侵信号类型的特征矢量; 最后在BPNN参比模型的基础上，用粒子群算法优化SVM的识别参数，使识别模型具有更强的自适应和自学习能力，克服了神经网络易陷入局部最优的不足之处，实现不同振动入侵信号的特征矢量识别。分析结果表明，针对不同类型的入侵源识别，该方法可以有效剔除随机噪声的影响，提取传感信息的特征矢量，降低异常值的影响，算法的预测类别与输出类别几乎一致，振动识别的精确率达到95%以上，识别效果明显强于BPNN网络的检测算法，提高了信息分析的准确性。

An Extraction and Recognition Method of the Distributed Optical Fiber Vibration Signal Based on EMD-AWPP and HOSA-SVM Algorithm

Given that the traditional signal processing methods can not effectively distinguish the different vibration intrusion sig-nal ,a feature extraction and recognition method of the vibration information is proposed based on EMD-AWPP and HOSA-SVM ,using for high precision signal recognition of distributed fiber optic intrusion detection system .When dealing with differ-ent types of vibration ,the method firstly utilizes the adaptive wavelet processing algorithm based on empirical mode decomposi-tion effect to reduce the abnormal value influence of sensing signal and improve the accuracy of signal feature extraction .Not only the low frequency part of the signal is decomposed ,but also the high frequency part the details of the signal disposed better by time-frequency localization process .Secondly ,it uses the bispectrum and bicoherence spectrum to accurately extract the feature vector which contains different types of intrusion vibration .Finally ,based on the BPNN reference model ,the recognition param-eters of SVM after the implementation of the particle swarm optimization can distinguish signals of different intrusion vibration , which endows the identification model stronger adaptive and self-learning ability .It overcomes the shortcomings ,such as easy to fall into local optimum .The simulation experiment results showed that this new method can effectively extract the feature vector of sensing information ,eliminate the influence of random noise and reduce the effects of outliers for different types of invasion source .The predicted category identifies with the output category and the accurate rate of vibration identification can reach above 95% .So it is better than BPNN recognition algorithm and improves the accuracy of the information analysis effectively .