摘要
提出一种基于对手建模的多飞行器协同拦截策略。首先,针对滑翔飞行器机动方式不确定、轨迹预测难度大的问题,采用贝叶斯神经网络进行对手建模,基于目标探测时序信息预估滑翔飞行器剩余飞行时间、末段机动高度、对抗成功率等。然后,建立多飞行器协同对抗仿真环境,设计基于模糊逻辑的拦截任务分配策略,采用多智能体近端策略优化(MAPPO)算法设计协同拦截制导律,并在多场景训练环境中进行元学习训练,进一步提高算法网络的泛化能力和鲁棒性。仿真结果表明:基于对手建模的多飞行器协同拦截策略相比多飞行器无协同、无对手建模的情况,可以充分发挥群体优势,最大程度围猎目标,提高群体拦截成功率。
A multi-vehicle cooperative interception strategy based on opponent modeling is proposed.Firstly,to address the challenges of uncertain maneuver modes and difficult trajectory prediction for glider vehicles,a Bayesian neural network is utilized to model the opponent.Based on the time series information of target detections,estimates are made for the remaining flight time,final maneuver altitude,and confrontation success rate of the glider vehicle.Next,a simulation environment for multi-vehicle cooperative confrontation is established.A fuzzy logic-based interception task assignment strategy is designed,and the multi-agent proximal policy optimization(MAPPO)algorithm is employed to devise the cooperative interception guidance law.To enhance the algorithm’s adaptability across various scenarios,a multi-scenario training environment is created.The network parameters of the MAPPO algorithm are used as initial parameters for meta-learning training,further improving the network’s generalization capability and robustness.Simulation results demonstrate that the multi-vehicle cooperative interception strategy based on opponent modeling can fully leverage the group’s advantages,effectively hunt the target,and improve the interception success rate compared to non-cooperative multi-vehicle interception modes without opponent modeling.
作者
惠耀洛
许波
李秀敏
孙均政
HUI Yaoluo;XU Bo;LI Xiumin;SUN Junzheng(Beijing System Design Institute of the Electro-mechanic Engineering,Beijing 100854,China)
出处
《宇航学报》
北大核心
2025年第3期601-615,共15页
Journal of Astronautics
关键词
协同拦截制导
多智能体
强化学习
剩余飞行时间
拦截任务分配
Cooperative interception and guidance
Multi-agent
Reinforcement learning
Time-to-go
Intercept task assignments
