基于两级SVM的Φ-OTDR系统振动事件识别算法

Vibration Event Recognition Algorithm of Φ-OTDR System Based on Two-Level Support Vector Machine

相位敏感光时域反射计(Φ-OTDR)对于振动信号能够实现高灵敏度的连续分布式测量,目前的振动事件识别算法常从一个或者两个维度来提取特征,如时域或频域等,未能实现多维度大样本特征参量的融合分析;现有的算法一般采用简单的单级识别算法,结构比较简单,导致最终的模型识别准确率不高、泛化能力较差。针对上述问题,对实验采集的振动信号从时域、频域和空间域的多参量特征进行提取和融合,针对具体的振动信号识别问题,构建了一种两级支持向量机(SVM)识别算法,对振动事件进行两级分类,能够实现对相似振动事件的精确识别,识别准确率达90%以上。

Phase sensitive optical time domain reflectometer(Φ-OTDR)can realize continuous and distributed measurement with high sensitivity for vibration signals, so it is often used for the identification and classification of vibration events. However, the current vibration event recognition algorithms often extract features from a single dimension, such as time domain or frequency domain, which can not achieve the fusion analysis of multi-dimensional and large sample feature parameters. At the same time, the existing algorithms generally use a simple unipolar recognition algorithm, which has a relatively simple structure, resulting in low accuracy and poor generalization ability of the final model recognition. To solve the above problems, multi parameter features of the experimental vibration signals are extracted and fused from the time domain, frequency domain and spatial domain. At the same time, for the specific vibration signal recognition problem, a two-level support vector machine(SVM)recognition algorithm is constructed to classify the vibration events, which can realize the accurate recognition of similar vibration events, and the recognition accuracy is more than 90%.