基于AdEMAMix优化算法的波前校正系统  

Wavefront Correction System Based on AdEMAMix Optimization Algorithm

作　　者：陈鹏 夏兵辉 CHEN Peng;XIA Binghui(Jinling Institute of Technology,Nanjing 211169,China)

机构地区：[1]金陵科技学院网络与通信工程学院,江苏南京211169

出　　处：《金陵科技学院学报》2025年第1期52-60,共9页Journal of Jinling Institute of Technology

基　　金：国家自然科学基金面上项目(12473081)。

摘　　要：针对常规随机并行梯度下降(SPGD)算法收敛速度慢、易陷入局部极值等问题,提出了将AdEMAMix优化算法与常规SPGD算法融合应用到自适应光学系统控制中,以提升波前校正效果。AdEMAMix优化算法引入快速和低速双指数移动平均机制,同时利用近期和远期的梯度信息进行梯度自适应,动态调节并优化过程中的增益系数,从而提高了收敛的稳定性和速度。在波前校正实验中,分别采用了不同单元的变形镜作为校正器件,模拟了在不同湍流强度条件下的波前像差校正。仿真结果显示:与传统SPGD算法相比,采用AdEMAMix优化算法在收敛速度和校正效率上表现更佳。此外,随着变形镜单元数和湍流强度的提高,AdEMAMix优化算法的优势更加明显。To address the problems of slow convergence speed and susceptibility to local optima in conventional stochastic parallel gradient descent(SPGD)algorithms,a fusion of AdEMAMix optimization algorithm and conventional SPGD algorithm is proposed for the control of adaptive optics system to improve the performance of wavefront correction.The AdEMAMix optimization algorithm introduces a fast and low-speed dual exponential moving average mechanism,while utilizing recent and long-term gradient information for gradient adaption,dynamically adjusting and optimizing the gain coefficient during the process to improve the stability and speed of convergence.In the wavefront correction experiments,deformable mirrors of different units are used as corrective devices to simulate wavefront aberration correction under different turbulence conditions.The simulation results show that compared with the conventional SPGD algorithm,the optimization algorithm with AdEMAMix performs better in terms of convergence speed and correction efficiency.In addition,as the number of deformable mirror units increases and the turbulence intensity strengthens,the advantages of the AdEMAMix optimization algorithm become more evident.

关 键 词：自适应光学系统 波前校正 SPGD算法 AdEMAMix优化算法 

分 类 号：TP273.2[自动化与计算机技术—检测技术与自动化装置]