基于CNN的光纤振动传感信号特征提取和识别

Feature Extraction and Recognition of Optic Fiber Vibration Sensing Signal Based on CNN

光纤振动传感系统极易受到噪声背景、多源干扰等影响,系统对目标事件检测识别的准确率直接决定了系统的运行效果,如何有效区分机械及人工动土作业信号和对管道威胁较小的环境干扰信号,在保证对动土作业行为有效报警的同时,排除非威胁性的环境干扰,提升系统的报警准确率,节省人力资源是需要重点解决的问题。对油气管道周边常见的威胁信号和非威胁信号进行了分析,并对基于机器学习模型的多维信号检测识别算法进行了探究。对CNN-时频图方法、1D-CNN和CNN-BiLSTM等三种思路进行了探究和比较。在准确率和计算耗时等方面,1D-CNN和CNN-BiLSTM方法更具优势。

The Fiber optic vibration sensing system is easily affected by noise background,multi-source interference,etc.The accuracy rate of the system's detection and identification of target events directly determines the operation effect of the system.How to effectively distinguish mechanical and manual earth-moving operation signals from environmental interference signals that pose less threat to pipelines?To improve the alarm accuracy of the system and save human resources is the key problem to be solved.The common threat signal and non-threat signal around oil and gas pipeline are analyzed,and the multi-dimensional signal detection and recognition algorithm based on machine learning model is explored.Three kinds of thinking,namely,CNN-time-frequency graph method,1D-CNN and CNN-BiLSTM,are explored and compared.In terms of accuracy and calculation time,1D-CNN and CNN-BiLSTM are more advantageous.