深度学习在飞行器动力学与控制中的应用研究综述

飞行器动力学与控制历经多年的发展取得辉煌成就,也面临着一系列亟需解决的难题。深度学习基于存储、记忆、预训练的新模式为动力学与控制难题的解决提供了新途径。本文以提升飞行器控制自主性和智能水平为研究主题,从三个方面总结了深度学习在飞行器动力学与控制中的应用,包括：在动力学建模中应用深度学习来提升模型计算效率和建模精度、求解模型反问题;在最优控制中应用深度学习来提升轨迹规划速度、最优控制实时性和自主性;在飞行器任务设计中应用深度学习来提升任务优化的计算效率和决策水平。在此基础上,梳理了各个方案的优缺点和部分代表性成果。最后,给出了深度学习应用于飞行器动力学与控制的四点建议。

A REVIEW ON THE APPLICATIONS OF DEEP LEARNING IN AIRCRAFT DYNAMICS AND CONTROL

The aircraft dynamics and control has achieved a significant progress, but also faces a series of problems that need to be dealt with. Deep learning provides a new solution for these problems, and it is good in many aspects, such as the working model of the experience storage, the intelligent accumulation and the off-line training. In this study, around the subject of the autonomy and the intelligence enhancement for the flight control, the applications of the deep learning in aircraft dynamics and control are reviewed in three aspects: (1) applications of deep learning in dynamic modeling to improve the computational efficiency and accuracy of modeling, or to solve the problem of inverse dynamics; (2) applications of deep learning in optimal control to improve the speed of trajectory planning or the real-time performance and autonomy of flight control; (3) applications of deep learning in mission design to improve optimization speed and decision-making intelligence. Furthermore, the advantages and the disadvantages are analyzed and representative papers are introduced. Finally, four suggestions to apply the deep learning in aircraft dynamics and control are given.