【摘要】 采用中国科学院合肥物质科学研究院自行研制的高精度激光大气传输四维数值模拟程序（HELP-4D），针对1μm波段固态激光在典型应用场景下的大气传输过程进行数值仿真，构建了40 000组数据集。以数据集内激光发射系统参数、传输场景参数、传输效应特征参数等作为输入，激光到达目标处光斑扩展半径、扩展倍数等光束质量评价因子作为输出，采用机器学习技术，建立了基于Lasso回归算法的评估模型。结果表明，当大气相干长度r₀（1～100 cm）、热畸变参数N_D（0～500）、大气透过率T（0～1）等特征参数变化幅度较大时，该模型均能较好地评估激光传输效果，模型评估结果与数据集中仿真数据的平均相对误差均≤5%，决定系数和相关系数均≥0.98，单次评估时间在微秒量级。该模型能够快速准确地评估激光大气传输到靶光束质量，为先进激光系统的实际应用提供技术支撑。更多还原

【Abstract】 Objective Laser atmospheric propagation is influenced by combined effects including turbulence, thermal blooming, atmospheric inhomogeneity, and other perturbations. Key beam quality metrics—such as target spot expansion ratio, spot radius growth, centroid displacement, and encircled energy ratio—quantify beam distortion and attenuation during atmospheric propagation, enabling systematic evaluation of laser propagation performance.Existing models fall into three categories: wave-optics models, empirical scaling-law models, and statistical analysis models. Wave-optics models provide high precision but suffer from prohibitive computational complexity for real-time applications. Empirical models simplify calculations but fail under extreme conditions(e.g., strong turbulence or thermal blooming). Statistical models enable rapid predictions but produce ensemble-averaged results insensitive to transient/local variations, require stringent data quality, and lack interpretability.This study introduces a Lasso regression-based framework to address these limitations, achieving real-time capability, high accuracy, and interpretability for laser atmospheric propagation assessment.Methods Figure 1 outlines the Lasso regression modeling workflow(Fig.1). Simulation data were generated using a four-dimensional high-energy laser atmospheric propagation and adaptive optics compensation code developed by the Anhui Institute of Optics and Fine Mechanics, Chinese Academy of Sciences(Tab.1). The code implements a multi-phase-screen propagation model, with datasets comprising laser parameters(wavelength,power), atmospheric parameters(turbulence strength, thermal blooming distortion), and beam quality metrics.Lasso regression with L1 regularization was applied to model beam quality degradation mechanisms,automatically selecting dominant features from high-dimensional data while suppressing noise. Hyperparameters(regularization strength, convergence tolerance) were optimized via grid search(Fig.2, Tab.2).Results and Discussions The Lasso regression-based model resolves critical limitations of conventional methods in real-time performance, accuracy, and feature interpretability(Tab.3). Leveraging Lasso’s feature selection mechanism, the model achieves precise predictions of beam quality metrics while maintaining computational efficiency and interpretability. Compared to traditional statistical models, it delivers superior prediction accuracy and faster computation, fulfilling real-time evaluation requirements in practical engineering scenarios. Simulation analyses demonstrate robust performance under complex atmospheric conditions, including s trong turbulence and thermal blooming(Fig.3-Fig.7).Conclusions The proposed Lasso regression model enables rapid, accurate evaluation of laser atmospheric propagation under extreme conditions, addressing the trade-off between computational cost and physical fidelity.Its embedded feature selection mechanism aligns with laser propagation physics(e.g., turbulence-driven beam wander vs. thermal blooming-induced defocus), enhancing interpretability for field deployment. Future efforts will extend the framework to multi-wavelength/pulse regimes and hybrid machine learning architectures(e.g.,physics-informed neural networks) for improved generalizability.更多还原