基于深度强化学习的飞行器过载和姿态智能控制研究  

Intelligent Control of Aircraft Overload and Attitude Based on Deep Reinforcement Learning

作　　者：谭富威 何永宁 孙晓晖 朱震 张庆昊 卢俊国[1] TAN Fuwei;HE Yongning;SUN Xiaohui;ZHU Zhen;ZHANG Qinghao;LU Junguo(School of Electronic Information and Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200240;Shanghai Aerospace Control Technology Institute,Shanghai 201109)

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240 [2]上海航天控制技术研究所,上海201109

出　　处：《飞控与探测》2025年第1期25-31,共7页Flight Control & Detection

基　　金：中国航天科技集团有限公司第八研究院产学研合作基金(USCAST2022-34)。

摘　　要：针对复杂多变环境下的飞行器过载和姿态智能控制问题,提出了一种基于柔性动作评价(Soft Actor-Critic,SAC)强化学习算法的分布式智能体控制算法,建立了分布式高效环境交互的深度强化学习算法框架和飞行器过载和姿态智能控制算法系统,增加了强化学习算法训练的数据量级和数据分布,提高了飞行器控制算法的性能和鲁棒性。通过在仿真环境中的实验结果表明,基于训练得到的智能体能够有效地在无人飞行器仿真过程中进行过载和姿态控制,分布式SAC算法在无人飞行器仿真场景中的控制效果优于原始SAC算法。This paper addresses the problem of intelligent control of aircraft overload and attitude in complex and changing environments.It proposes a distributed intelligent agent control method based on the Soft Actor-Critic(SAC)algorithm,establishes a framework of distributed efficient environment interaction for deep reinforcement learning algorithms,and designs an intelligent control algorithm system for aircraft overloadand attitude.This approach increases the scale and distribution of training data for reinforcement learning algorithms,thereby improving the performance and robustness of aircraft control algorithms.Experimental results in simulated environments demonstrate that the trained intelligent agents effectively control overloadsandattitudes in unmanned aerial vehicle simulations.The distributed SAC algorithm outperforms the original SAC algorithm in controlling unmanned aerial vehicles in simulation scenarios.

关 键 词：深度强化学习 无人飞行器 分布式SAC算法 过载控制 姿态控制 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]