航空学报 > 2020, Vol. 41 Issue (6): 523467-523467   doi: 10.7527/S1000-6893.2019.23467

战斗机嵌入式训练系统中的智能虚拟陪练

陈斌, 王江, 王阳   

  1. 中国航空工业成都飞机设计研究所, 成都 610074
  • 收稿日期:2019-09-08 修回日期:2019-10-08 出版日期:2020-06-15 发布日期:2019-11-14
  • 通讯作者: 陈斌 E-mail:cb_cy2011@163.com

Intelligent virtual training partner in embedded training system of fighter

CHEN Bin, WANG Jiang, WANG Yang   

  1. AVIC Chengdu Aircraft Design and Research Institute, Chengdu 610074, China
  • Received:2019-09-08 Revised:2019-10-08 Online:2020-06-15 Published:2019-11-14

摘要: 智能化"实虚"对抗是现代先进战斗机嵌入式训练系统的重要功能需求。自主空战决策控制技术在未来空战装备发展中扮演关键角色。将当前的功能需求和发展中的技术结合起来,得到了空战智能虚拟陪练的概念。先进控制决策技术的引入使得智能虚拟陪练能够帮助飞行员完成复杂的战术训练,而训练中真实的对抗场景为技术的验证提供了理想的环境,大量的训练数据为技术的持续迭代优化提供了保障。作为可学习和进化的空战战术专家,智能陪练在人机对抗和自我对抗中不断优化,当其具备与人相当甚至超越人的战术能力时,可应用于未来的无人空战系统。智能虚拟陪练需要具备4项基本能力:智能决策能力、知识学习能力、对抗自优化能力和参数化表示能力。对其包含的关键技术进行了分析,提出并实现了一个基于模糊推理、神经网络和强化学习的解决方案,展示了其各项基本能力及目前达到的空战水平。未来更多的模型和算法可在智能虚拟陪练的框架中进行验证和优化。

关键词: 嵌入式训练系统, 智能虚拟陪练, 自主空战, 模糊推理, 神经网络, 强化学习

Abstract: Intelligent ‘live vs. virtual’ counterwork is an important function requirement for modern advanced fighter’s embedded training system. Autonomous decision making and control technology plays a vital role in the development of future air battle equipment. The combination of the current requirement and developing technology creates the concept of air combat intelligent virtual training partner. Advanced decision-making and control technology enables the intelligent virtual training partner to help pilots fulfill complex tactics training, during which live scenarios and amount of data provide ideal validation environment and continuous optimization opportunities. As an air combat tactics expert that allows self-learning and self-evolving, the intelligent training partner is able to get consistent optimization in counterwork with pilots and itself. And it can be applied into future unmanned air battle systems when it has tactical advantages equal even superior to pilots. The four basic capability requirements for intelligent virtual training partner are presented: intelligent decision making, knowledge learning, self-countering and optimizing, and parameterization representing. The key technologies involved in these requirements are analyzed. A prototype solution is built based on fuzzy inference, neural network and reinforcement learning, and their basic capabilities and current air combat level were shown in this paper. In the future, more models and algorithms can be validated and optimized in the framework of intelligent virtual training partner.

Key words: embedded training system, intelligent virtual training partner, autonomous air combat, fuzzy inference, neural network, reinforcement learning

中图分类号: