计量学报2025，Vol.46Issue(6)：853-861,9.DOI:10.3969/j.issn.1000-1158.2025.06.10
基于神经网络的BFS误差估算方案及影响因素研究
Estimation Scheme and Influence Factors Analysis of BFS Error Based on Artificial Neural Network
摘要
Abstract
With the development of artificial intelligence technology,artificial neural network(ANN)are widely used to extract Brillouin frequency shift(BFS)from Brillouin spectrum.Comparing ANN with linear activation function(ANN-P),ANN with nonlinear activation function,and classical spectral fitting methods,it was found that ANN-P has similar accuracy to nonlinear ANN,but requires less training time.Its accuracy is comparable to classical spectral fitting algorithms,with a computation time of only 0.70%of the Lorentz spectral fitting algorithm and 0.43%of the pseudo-Voigt spectral fitting algorithm.Through theoretical derivation,a BFS error estimation scheme based on ANN-P is derived.Based on this scheme,Brillouin gain error,weight matrix,and sweep frequency range,BFS error can be quickly estimated.The influence of line width,signal-to-noise ratio,and number of sweep frequency points on BFS error is also derived.The correctness of the scheme and the rules are verified through the calculation of a large amount of Brillouin data with different linewidths,signal-to-noise ratios,and scanning frequency points generated and measured numerically.For the measured spectrum,the proposed BFS error estimation scheme has an average absolute error of 7.99×10-2 MHz and an average relative error of 15.87%.关键词
光学计量/分布式光纤传感/人工神经网络/布里渊频移/误差估算Key words
optical metrology/distributed fiber sensing/artificial neural network/Brillouin frequency shift/error estimation分类
通用工业技术引用本文复制引用
赵丽娟,陈永辉,徐志钮,张旭哲..基于神经网络的BFS误差估算方案及影响因素研究[J].计量学报,2025,46(6):853-861,9.基金项目
国家自然科学基金(62273146,62171185) （62273146,62171185）
河北省自然科学基金(E2020502010) （E2020502010）
河北省省级科技计划(SZX2020034) （SZX2020034）
