基于深度强化学习的尾旋改出技术  

Aircraft Spin Recovery Technique Based on Deep Reinforcement Learning

作　　者：谭健美 王君秋 Tan Jianmei;Wang Junqiu(Chinese Aeronautical Establishment,Beijing 100029,China)

机构地区：[1]中国航空研究院,北京100029

出　　处：《航空兵器》2024年第1期77-88,共12页Aero Weaponry

摘　　要：本文搭建了飞机仿真环境,基于近端策略优化(PPO)算法建立了尾旋改出算法测试模型,设计了基准版单阶段、基准版双阶段、加深版单阶段、加深版双阶段四种网络结构,用于探究网络结构和改出阶段对尾旋改出效果的影响,设置了鲁棒性测试试验,从时延、误差和高度等方面进行了算法测试和结果分析。This paper builds an aircraft simulation environment,and establishes a test model of an automated spin recovery algorithm based on proximal policy optimization(PPO)algorithm.Four kinds of network structures are designed,that are basis single stage,basis double stage,deep single stage and deep double stage,to explore the influence of network structure and recovery stage on spin recovery effect.A robustness test experiment is set up,and the algorithm is tested and the results are analyzed from the aspects of delay,error and height.

关 键 词：尾旋改出 深度学习 强化学习 近端策略优化 算法测试 飞机 

分 类 号：TJ760[兵器科学与技术—武器系统与运用工程] V271.4[航空宇航科学与技术—飞行器设计]