为了对分布式光纤上的入侵信号类型进行准确识别,提出了一种基于集合经验模态分解(EEMD)结合随机向量函数链接(RVFL)神经网络的光纤入侵信号的特征提取与识别算法。算法步骤为:对采集到的光纤入侵信号作预处理操作,包括最小-最大规范化处理和利用db3小波去除信号的低频噪声;采用EEMD方法对入侵信号进行分解,得到5组本征模态函数(IMF);计算各IMF分量的能量占比,并依据方差分析法筛选出3组特征向量;将特征向量送入RVFL神经网络进行训练并对入侵信号进行识别。实验结果显示:该方法能正确识别不同入侵信号的类型,具有较高的准确率。

在光纤振动信号检测系统中，使用相位敏感光时域反射技术，可检测长距离的光纤振动信号，但反射回的光信号都有大量的振动杂波和噪声，导致振动检测信号的虚警率较高。提出采用单元平均恒虚警(CA-CFAR)和有序统计恒虚警(OS-CFAR)二级检测算法，保持虚警率稳定，既提高了运算速度，也改善了参考单元中存在多个目标的检测性能，并提出利用蒙特卡罗方法确定二级检测的门限系数，最后通过蒙特卡罗仿真及现场实验验证，对算法进行了性能分析，验证了算法的可行性以及有效性。

One of the key technologies for optical fiber vibration pre-warning system (OFVWS) refers to identifying the vibration source accurately from the detected vibration signals. Because of many kinds of vibration sources and complex geological structures, the implement of identifying vibration sources presents some interesting challenges which need to be overcome in order to achieve acceptable performance. This paper mainly conducts on the time domain and frequency domain analysis of the vibration signals detected by the OFVWS and establishes attribute feature models including an energy information entropy model to identify raindrop vibration source and a fundamental frequency model to distinguish the construction machine and train or car passing by. Test results show that the design and selection of the feature model are reasonable, and the rate of identification is good.