基于深度强化学习的三体对抗博弈策略研究

Three-body adversarial game strategies based on deep reinforcement learning

针对三体对抗场景中的攻防博弈问题,提出了基于深度强化学习的智能博弈策略,包括适用于进攻弹的攻击策略以及适用于目标/防御弹的主动防御策略。在经典三体对抗研究的基础上引入强化学习算法,提高了算法训练的目的性,同时在奖励函数设计中考虑了攻防对抗双方的奖惩条件。应用深度强化学习算法对攻防双方智能体进行训练,并得到收敛的博弈策略。仿真结果表明,通过训练获得的进攻弹的攻击策略能够根据战场态势合理规划机动行为,在避开防御弹攻击后仍能在短时间内成功命中目标;目标/防御弹的主动防御策略中的目标扮演诱饵角色,防御弹将进攻弹迅速锁定在拦截三角形上,从而使目标在战场上面临机动能力较强的进攻弹时,能够免于攻击。

Aiming at the attack and defense game problem in the three-body confrontation scenario,an intelligent game strategy based on deep reinforcement learning is proposed,including attack strategies applicable for attacking missiles and active defense strategies for target/defender.Based on the classical threebody adversarial research,the reinforcement learning algorithm is introduced to improve the purposefulness of the algorithm training,while the reward and punishment conditions of both attack and defense confrontation are considered in the reward function design.A deep reinforcement learning algorithm is applied to train the agents,and a convergent game strategy is obtained.The simulation results show that the strategy of the attacking missile obtained from training can reasonably plan the maneuvering behavior according to the battlefield situation,and can successfully hit the target in a short time while avoiding the attack of the defender.The target in the active defense strategy of target/defender plays the role of a decoy,and the defender locks the attacking missile onto the intercept triangle quickly,thus making the target immune to attack as facing the attacking missile with superior maneuvering capability on the battlefield.