基于灰狼算法优化的1D CNN用于分布式光纤传感扰动事件识别  

Distributed Fiber Sensing Disturbance Event Recognition Based on GWO Optimizing 1D CNN

作　　者：晏晨 张大路 吴海勇 陈勐勐 YAN Chen;ZHANG Dalu;WU Haiyong;CHEN Mengmeng(School of Information Engineering,Yancheng Institute of Technology,Yancheng 224051,China;School of Electronic Engineering,NanJing XiaoZhuang University,Nanjing 211171,China)

机构地区：[1]盐城工学院完信息工程学院,江苏盐城224051 [2]南京晓庄学院电子工程学院,江苏南京211171

出　　处：《量子光学学报》2025年第1期42-52,共11页Journal of Quantum Optics

基　　金：国家自然科学基金(62105157);江苏省自然科学基金(2021K228B);江苏省博士后基金(BK20211014)。

摘　　要：使用卷积神经网络(CNN,Convolutional Neural Network)来对分布式光纤传感系统所采集到的扰动信号进行识别的方法已经非常常见。该方法借助卷积神经网络强大的特征提取能力,相比于传统的机器学习方法,无需手动提取事件信号中的特征。但是神经网络避免不了调参的环节,于是本文提出了一种改进的灰狼算法(GWO,Gray Wolf Optimizer)来实现对网络参数的自动调节。灰狼算法最重要的是目标函数的选取,本文以训练集上的准确率作为目标函数,将神经网络训练过程中的卷积核大小、批处理数量以及每一个卷积层和第一个全连接层的输出维度作为待优化的参数,这些参数不断迭代从而尽可能地找到目标函数取最大值时所对应的参数组合。训练结果显示,卷积神经网络使用通过改进后的灰狼算法优化的参数在测试集上的识别准确率可达96%,而优化前的准确率为94%,说明改进后的灰狼算法用于参数优化确实可以提高神经网络训练的准确性。Objective The method of using Convolutional Neural Network(CNN)for identifying disturbance signals which acquired from distributed fiber optic sensing systems has become quite common.Based on the tremendous success in image processing,Convolutional Neural Network has widely applied to the analysis and feature extraction of time-series data.In the distributed fiber optic sensing systems,disturbance signals are often caused by various factors,leading to complex and varied patterns.Traditional machine learning methods usually require experts to manually design feature extractors based on experience,which is not only time-consuming but also has potentially biased,affecting the final identification results.Compared with machine learning methods,although Convolutional Neural Network can automatically learn deep features from the data,greatly simplifying the feature extraction process,but their performance in practical applications is largely determined by the configuration of parameters.Adjusting these parameters is typically a trial-and-error process that requires substantial time and computational resources.To address this issue,this paper proposes an improved Gray Wolf Optimizer(GWO)algorithm for the automated optimization of Convolutional Neural Network parameters.Methods Gray Wolf Optimizer is a intelligent optimization algorithms that simulates the hunting behavior of gray wolves.It has the advantages of simple structure,fast convergence speed and strong global search ability.The most important aspect of the Gray Wolf Optimizer is the selection of the objective function.In this paper,the accuracy on the training set is used as the objective function,and the convolution kernel size,batch size,and output dimensions of each convolutional layer and the first fully connected layer during the training process of the neural network are taken as the parameters to be optimized.These parameters are iteratively adjusted to find the parameter combination that maximizes the objective function as much as possible.Results and D

关 键 词：分布式光纤传感技术 灰狼算法 卷积神经网络 扰动信号识别 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置]