中国科学院自动化研究所机构知识库

近年来，随着无人机对抗呈现多任务、多目标等特点，多机协同对抗受到广泛关注。但现阶段无人机自主能力较低，还不能完全替代有人机作业，因此各国对多无人机自主协同对抗的探索和研究仍在不断继续。本文以超视距中距空战为研究背景，开展多无人机协同对抗系统的智能决策和控制研究。主要工作如下：

（1）无人机建模与飞行控制系统设计：根据无人机的运动特性，建立了无人机质点模型；设计了包括速度控制子系统、高度控制子系统和航迹方位角控制子系统的无人机飞行控制系统，实现了对无人机质点飞行的平稳快速控制。

（2）基于min-max矩阵博弈的无人机对抗决策方法设计：本文将无人机自身性能与空战实际战术进行了总结和简化处理，构建了专家动作空间；利用非参量法，构建了具有超视距中距空战特点的态势评估函数；基于min-max矩阵博弈设计了无人机自主对抗的决策方法，使无人机在对抗中具有实时战术生成和在线决策的自主能力。

（3）基于单一输入规则群（Single Input Rule Modules, SIRMs)动态加权模糊推理模型与改进型自适应遗传算法（Improved Adaptive Genetic Algorithm, IAGA)的多机协同对抗决策方法研究：根据SIRMs推理框架，本文为每个影响决策的输入变量设计了SIRM与动态权重，使得复杂的输入变量被解耦，同时可使不同输入对决策产生不同影响；此外，利用IAGA优化模糊规则，在IAGA中，同时考虑对抗结果和过程，基于有序时间长度上的态势设计了IAGA的适应值函数，设计了精英策略选择、自适应单点交叉和退火变异，实现了仅在简单专家知识框架的支持下即可得到不同对抗环境中的精细化推理规则的目标。

（4）多无人机协同对抗仿真系统开发：实现了多机对抗过程的可视化，为多无人机协同对抗系统提供了简单高效的开发与验证环境。此外，结合项目需求，设计了面向小型移动机器人（无人车）的集群验证系统，完成了预规划、专家规则等算法的验证实验。该系统可为后续进行本文的决策算法验证提供支持。

With the multi-task and multi-target characteristics of UCAVs confrontation, the issue of multiple unmanned combat aerial vehicles (UCAVs) cooperative air combat has attracted widespread attention in recent years. However, the autonomous capability of UCAVs is being at a low level. The UCAV struggles to replace manned aircraft at present. Therefore, the exploration and research on the autonomous air combat of multiple UCAVs are ongoing in many countries. Based on the beyond-visual-range middle range air combat, this thesis focuses on intelligent tactical decision and control of multiple UCAVs cooperative countermeasure system.

It contributes to (1) design the model and flight control system of the UCAV. According to the kinematic characteristics of the UACV, model of the UCAV mass point is established. In addition, flight control system of the UCAV, which makes the UCAV achieve smooth control in particle flight, is designed including speed, altitude and flight path azimuth control subsystem.

(2) Design a tactical decision method of UCAVs based on min-max matrix game principle. In the thesis, the performance and the actual tactics of UCAVs are summarized and simplifized, which make the expert action space constructed. Furthermore, based on the nonparametric method, a situation assessment function with characteristics of beyond-visual-range middle range air combat is built. Finally, based on the min-max matrix game method, the decision-making method of UACVs autonomous countermeasure is designed, which realizes UACVs autonomous capability of real-time tactical generation and online decision-making in combat.

(3) Propose a tactical decision method based on Single Input Rule Modules (SIRMs) dynamically connected fuzzy inference model and improved adaptive genetic algorithm (IAGA) for multiple UCAVs air-to-air combat. According to the SIRMs inference framework, a SIRM and dynamic important degree for each input variable are designed. The complex input variables are decoupled, and at the same time the different effects of different inputs on decision-making can be reflected by this design. In addition, IAGA is adopted to optimize fuzzy rules. In IAGA, air combat results and process are combined to design the fitness value function based on the situation assessment values over the orderly time length. At the same time, adaptive genetic algorithm is improved, in which the elite strategy selection method, a single-point crossover method with adaptive crossover probability and the annealing mutation based on simulated annealing ideas are adopted. With a simple rule skeleton, the optimal detailed rule can be produced in different combat environment by continuous iterative optimization.

(4) Develop a multiple UCAVs cooperative countermeasure simulation systems to realize the visualization of the multiple UCAVs countermeasure process. The system provides a simple and efficient development and verification environment for the multiple UCAVs cooperative countermeasure system. In addition, in combination with the actual project requirements, cluster authentication system for small mobile robots (Unmanned Vehicles) is designed and implemented, and verification experiments of pre-planning and expert rules algorithms in the physical environment have been completed. The system can prepare for the support of the subsequent decision algorithm verification of this thesis.