The complexity and uncertainty of actual air combat and the unknown information of some air combat bring great challenges to unmanned aerial vehicle (UAV) air combat target intention prediction. In this paper, we examine the problem of air combat intention prediction under incomplete information, and present an air combat target intention prediction model based on longshortterm memory (LSTM) with incomplete information. The model adopts a hierarchical method to establish the feature set of air combat target intention prediction, encodes the information of air combat to time series features, encapsulates expert knowledge into labels, and introduces the method of fitting cubic sample interpolation function and filling average value to repair incomplete data. Also, we used the adaptive moment estimation (Adam) optimization algorithm to accelerate the training speed of the model to effectively prevent local optimum. Finally, the simulation results show that the proposed model can effectively predict the target intention of UAVs in air combat.
装备预研中国电科联合基金开放课题(6141B08231110a)
装备预研重点实验室基金项目(61425040104)
[1] Han P. Cooperative task planning technology for multiUAVs. Dissertation for Master Degree. Nanjing: Nanjing University of Aeronautics and Astronautics, 2013. Google Scholar
[2] Zhou T L, Chen M, Chen S D, et al. Intention prediction of aerial target under incomplete information. ICIC Express Letters, 2017, 8(3): 623631. Google Scholar
[3] Xia X. The study of target intent assessment method based on the templatematching. Dissertation for Master Degree. Changsha: Graduate School of National University of Defense Technology, 2006. Google Scholar
[4] Ge X, Xia X Z. DSBN used for recognition of tactical intention. Syst Eng Electron, 2014, 36: 7683. Google Scholar
[5] Wu Z Q, Li D F. A model for aerial target attacking intention judgment based on reasoning and multiattribute decision making. Electron Opt Control, 2010, 17: 1013. Google Scholar
[6] Cui Y P, Wu Q X, Chen M. Aerial target intention prediction based on adaptive neurofuzzy inference system. In: Papers of the 15th China Symposium on System Simulation Technology and Its Application. Hefei: China University of Science and Technology Press, 2014: 277281. Google Scholar
[7] Chen H, Reng Q L, Hua Y, et al. Fuzzy neural network based tactical intention recognition for sea targets. Syst Eng Electron, 2016, 38: 18471853. Google Scholar
[8] Zhou W W, Yao P Y, Zhang J Y, et al. Combat intention recognition for aerial targets based on deep neural network. Acta Aeronaut ET Astronaut Sin, 2018, 39: 200208. Google Scholar
[9] Ou W, Liu S J, He Y Y, et al. Tactical intention recognition algorithm based on encoded temporal features. Command Control Simul, 2016, 38: 3641. Google Scholar
[10] Ou W, Liu S J, He Y Y, et al. Study on the intelligent recognition model of enemy target's tactical intention on battlefield. Comput Simul, 2017, 34: 1014. Google Scholar
[11] Shun Z Z, Wu H W, Yuan W P, et al. Computational Method and Practice. Nanjing: Southeast University Press, 2011. Google Scholar
[12] Kingma D P, Ba J. Adam: a method for stochastic optimization. 2014,. arXiv Google Scholar
[13] Song Y, Zhang X H, Guo H D. Hierarchical inference frame and realization of air target tactical intention. Inform Command Control Syst Simul Technol, 2005, 27: 6366. Google Scholar
[14] Zhu B L. Effectiveness Evaluation of Combat Aircraft. Beijing: Aviation Industry Press, 1993. Google Scholar
[15] Hu B K. Fighter onetoone air combat simulation system and its application. Aviat Syst Eng, 1992, 5: 3543. Google Scholar
[16] Zhu K Q, Dong Y F. Research on design method of air combat maneuvering action base. Aeronaut Comput Technol, 2001, 31: 5052. Google Scholar
Figure 1
Intention representation and prediction process of air combat target
Figure 2
Intention prediction workflow of UAV air combat
Figure 3
(Color online) Air combat situation. (a) Advantage of our side; (b) advantage of the enemy;protect łinebreak (c) neutrality between the two sides; (d) balance of power between the two sides
Figure 4
Characteristic description tree chart of UAV air combat intention prediction
Figure 5
Framework of UAV air combat intention prediction model
Figure 6
Flow chart of air combat data repairing
Figure 7
Intention coding and pattern analysis
Figure 8
Fitting curve of distance in the case that some distance values are missing at some time points
Figure 9
(Color online) Recognition accuracy of training set and test set
Figure 10
(Color online) Comparison of different recognition models
Air combat situation  Most possible target intention  Secondary possible operational intention 
Advantage of the enemy  Attack  Surveillance 
Advantage of our side  Defense  Penetration, electronic interference 
Balance of power between the two sides  Feint  Attack, defense, electronic interference 
Neutrality between the two sides  Reconnaissance  Electronic interference 
初始化参数$\theta~$, 初始化一阶和二阶矩变量${m_0}$, ${v_0}$, 初始化时间步长$t~=~0$; 

从训练集中采包含$m$个样本$X' = \left\{ {{x_1},{x_2}, \ldots ,{x_m}} \right\}$和其对应的${\hat y_{x'}}$标签集合; 
计算梯度: ${g_{t~+~1}}~=~\frac{1}{m}{\nabla~_\theta~}\sum\nolimits_i~{L\left(~{f\left(~{{x_i};{\theta~_t}}~\right),{y_{{x_i}}}}~\right)}~$; 
$t~=~t~+~1$; 
更新有偏一阶矩估计: ${m_t}{\rm{~=~}}{\beta~_1}~\cdot~{m_{t~~1}}~+~\left(~{1~~{\beta~_1}}~\right)~\cdot~{g_t}$; 
更新有偏二阶矩估计: ${v_t}~=~{\beta~_2}~\cdot~{v_{t~~1}}~+~\left(~{1~~{\beta~_2}}~\right)~\cdot~g_t^2$; 
修正一阶矩的偏差: ${\hat~m_t}~=~{m_t}/\left(~{1~~\beta~_1^t}~\right)$; 
修正二阶矩的偏差: ${\hat~v_t}~=~{v_t}/\left(~{1~~\beta~_2^t}~\right)$; 
更新参数: ${\theta~_t}~=~{\theta~_{t~~1}}~~\alpha~~\cdot~{\hat~m_t}/\left(~{\sqrt~{{{\hat~v}_t}}~~+~\varepsilon~}~\right)$; 

Time point  Distance between two sides (km)  Time point  Distance between two sides (km)  Time point  Distance between two sides (km) 
1  6.14  5  4.76  9  5.77 
2  5.90  6  4.67  10  5.88 
3  5.72  7  4.97  11  6.08 
4  5.32  8  5.47  12  6.25 
Data missing  Model recognition  Data missing  Model recognition 
0  94.12  40  84.85 
10  93.53  50  73.76 
20  92.78  60  60.57 
30  90.69  70  46.67 
Recognition rate of  Test recognition  Recognition rate of  Test recognition 
training set  rate  training set  rate 
98.16  94.30  98.62  93.95 
98.12  93.91  98.06  93.89 
98.54  94.11  98.10  94.06 
98.57  93.91  98.12  94.26 
98.47  94.12  98.42  94.21 
Copyright 2020 CHINA SCIENCE PUBLISHING & MEDIA LTD. 中国科技出版传媒股份有限公司 版权所有
京ICP备14028887号23 京公网安备11010102003388号