高斯光束湍流大气传输远场扩展半径快速计算方法研究  

Study of fast calculation method of far-field radius for laser propagation through turbulent atmosphere

作　　者：崔询 陈小威 钱仙妹[2,3,4] 朱文越 武鹏飞[2,3,4] 闵令云 饶瑞中 CUI Xun;CHEN Xiaowei;QIAN Xianmei;ZHU Wenyue;WU Pengfei;MIN Lingyun;RAO Ruizhong(School of Environmental Science and Optoelectronic Technology,University of Science and Technology of China,Hefei 230026,China;Key Laboratory of Atmospheric Optics,Anhui Institute of Optics and Fine Mechanics,Hefei Institutes of Physical Science,Chinese Academy of Sciences,Hefei 230031,China;Anhui Laboratory of Advanced Laser Technology,Hefei 230037,China;Nanhu Laser Laboratory,National University of Defense Technology,Changsha 410073,China)

机构地区：[1]中国科学技术大学环境科学与光电技术学院,安徽合肥230026 [2]中国科学院合肥物质科学研究院安徽光学精密机械研究所中国科学院大气光学重点实验室,安徽合肥230031 [3]先进激光技术安徽省实验室,安徽合肥230037 [4]国防科技大学南湖之光实验室,湖南长沙410073

出　　处：《红外与激光工程》2024年第9期284-293,共10页Infrared and Laser Engineering

基　　金：脉冲功率激光技术国家重点实验室主任基金项目(SKL2020ZR08)。

摘　　要：高斯光束湍流大气传输的远场有效半径是工程中较为关注的远场光斑特征,文中研究其快速计算方法。基于多场景下高斯光束湍流大气传输波动光学仿真数据,建立了定标规律模型和3种人工智能模型LightGBM(Light Gradient Boosting Machine)、MLP(Multilayer Perceptron)、MFTT(Modified Feature Tokenizer Transformer),比较了不同模型的精度、泛化能力、样本量的影响和计算速度。结果表明:4种模型的精度受训练集样本量的影响,小样本量下定标规律模型精度最高,大样本量下MFTT模型精度最高;数据分为70%的训练集和30%的测试集时,MFTT在测试集上的平均相对偏差为1.36%;MFTT泛化能力最强,定标规律模型计算速度最快。希望为激光大气传输评估快速计算方法的选择和应用提供参考。Objective Laser beams propagating in turbulent atmosphere suffer from various factors such as diffraction,turbulence,aberration and jitter,which broaden the beam radius and significantly affect laser applications including laser propagation,communication and imaging.In recent years,researchers have made efforts to develop fast and high-precision methods for calculating far-field effective radius.Scaling laws and artificial intelligence models are commonly used for fast evaluation of laser beam propagation.However,comparison between these two methods is rarely reported.In this paper,the fast calculation method of far-field radius is studied,and the performance of the different models is compared in many aspects,hoping to provide reference for the selection and application of fast calculation methods for evaluation of laser atmospheric propagation.Methods Firstly,numerical simulation is carried out by using wave optical program,and the scaling index of the scaling law is determined by genetic algorithm.Secondly,three AI(artificial intelligence)models,MLP,LightGBM and FTT,are selected to evaluate the effective radius of the far field,Additionally,two specific improvements are made to the FTT model(Fig.2).Then,the hyperparameters of AI models are determined by using TPE algorithm(Fig.3-4).Finally,three artificial intelligence models are constructed to evaluate the far-field effective radius(Fig.5).Results and Discussions When the data is divided into 70%training set and 30%test set,the accuracy of the scaling law model and the three AI models are compared in the test set(Fig.1,Fig.6).The results show that the accuracy of the three artificial intelligence models is higher than that of the scaling law model,and the MFTT model has the highest accuracy with mean relative error of 1.36%.The effect of training set sample size on the evaluation accuracy of 4 models was studied(Fig.7).From the perspective of changing trend,the scaling law model is relatively stable,while the three AI models change sharply,which indicates tha

关 键 词：大气传输 人工智能 定标规律 精度 

分 类 号：O439[机械工程—光学工程]