仿生算法优化神经网络的光纤传感器光强补偿方法  

Optical fiber sensor intensity compensation method based on bionic algorithm optimizing neural network

作　　者：路茵 杨瑞峰 郭晨霞 LU Yin;YANG Ruifeng;GUO Chenxia(Automatic Test Equipment and System Engineering Research Center of Shanxi Province,School of Instrument and Electronics,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学仪器与电子学院,山西省自动化检测装备与系统工程技术研究中心,太原030051

出　　处：《激光杂志》2023年第9期150-154,共5页Laser Journal

基　　金：山西省中央引导地方科技发展自由探索类基础研究项目(No.YDZJSX2022A027)。

摘　　要：为了减少非线性因素对反射式光纤传感器测量结果的影响,提出了一种基于麻雀搜索算法(SSA)优化反向传播神经网络(BPNN)的光强补偿模型。利用麻雀搜索算法优化反向传播神经网络的权值与阈值,达到避免传统反向传播神经网络陷入局部样本极值的目的。比较麻雀搜索算法和粒子群算法的收敛曲线,可知麻雀搜索算法具有迭代步骤少和收敛速度快的特性。通过传感器内圈和外圈两组接收光功率值训练SSA-BP神经网络,结果表明该混合算法平均绝对误差为0.002,均方根误差为0.003,平均绝对百分比误差为0.011%。将SSA-BP神经网络、PSO-BP神经网络和传统BP神经网络的补偿效果进行对比,结果证明提出的光强补偿模型预测误差更小且稳定,能够更高精度完成位移测量过程。In order to reduce the influence of nonlinear factors on the measurement results of reflective optical fiber sensors,a light intensity compensation model based on Sparrow Search Algorithm(SSA)optimized Back Propagation Neural Network(BPNN)is proposed.The weights and thresholds of BPNN are optimized by SSA to avoid the traditional BPNN falling into local sample extremum.Comparing the convergence curve of the SSA and the particle swarm optimization algorithm(PSO),it can be seen that the SSA has the characteristics of less iterative steps and fast convergence speed.The SSA-BP neural network is trained by two groups of received light power values of the inner and outer rings of the sensor,the results show that the average absolute error is 0.002,the root mean square error is 0.003 and the mean absolute percentage error is 0.011%.Comparing the compensation effects of SSA-BP neural network with PSO-BP neural network and traditional BP neural network,it is proved that the proposed intensity compensation model has smaller and more stable prediction error and can complete the displacement measurement process with higher accuracy.

关 键 词：光纤传感器 仿生算法 BP神经网路 SSA-BP神经网络 光强补偿 

分 类 号：TN911[电子电信—通信与信息系统]