基于RF-XGBoost算法的无人机多回合攻防博弈决策  

Multi-round attack and defense game decision-making of UAVs based onRF-XGBoost algorithm

作　　者：邹世培 王玉惠[1] 刘鸿睿 ZOU Shipei;WANG Yuhui;LIU Hongrui(College of Automation Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China)

机构地区：[1]南京航空航天大学自动化学院,江苏南京211106

出　　处：《系统工程与电子技术》2025年第2期518-526,共9页Systems Engineering and Electronics

基　　金：科技创新2030“新一代人工智能”科技部国家重点研发计划(2018AAA0100805);前瞻布局科研专项项目(ILA220591A22)资助课题。

摘　　要：为解决不平衡空战数据集下的无人机多回合博弈对抗问题,提出一种随机森林-极限梯度提升(random forest-eXtreme gradient boosting, RF-XGBoost)算法以进行攻防博弈决策研究。通过分析红蓝双方的运动状态和空战信息,建立支付矩阵模型,利用线性归纳法求解当前博弈纳什均衡解和期望收益,以蓝方最终获胜作为博弈对抗是否停止的判断条件。在博弈对抗过程中,首先基于随机森林(random forest, RF)算法对空战数据集进行特征降维以提高空战决策的实时性,然后提出改进的XGBoost算法来处理不平衡数据集,将其用于确定最优机动动作以提高机动决策准确率和提升蓝方对抗态势,并得到下一回合的红蓝空战信息;之后,根据下一回合的支付矩阵模型重新计算纳什均衡解和期望收益,直至蓝方获胜;最后,通过仿真验证所提算法的可行性和有效性。To solve the multi-round game confrontation problem of unmanned aerial vehicles(UAVs)with unbalanced air combat data set,a random forest-eXtreme gradient boosting(RF-XGBoost)algorithm is proposed to study the attack and defense game decision-making.The payment matrix model is established by analyzing the movement status and air combat information of the red and blue sides,then,the linear induction method is considered to solve the current Nash equilibrium solution and expected return of the game,and whether the game confrontation will stop depends on the victory of the blue side.In the process of game confrontation,in the first place,the feature dimensionality reduction of air combat data set is conducted based on the random forest(RF)algorithm to improve the real-time performance of air combat decision-making.Then,an improved XGBoost algorithm is proposed to deal with the unbalanced data set,which is used to determine the optimal maneuvers to improve the accuracy of maneuver decision-making and enhance blue confrontation’s situation,and air combat information of the next round of red and blue sides is obtained.Furthermore,the Nash equilibrium solution and expected return based on the payment matrix model of the next round can be obtained once again,until the blue side wins.Finally,the feasibility and effectiveness of the proposed algorithm are verified by simulation test.

关 键 词：无人机 随机森林 极限梯度提升 多回合博弈 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]