演化博弈强化学习模型在智能博弈对抗中的应用

Application of Evolutionary Game Reinforcement Learning Model to Intelligent Gaming

智能化指挥控制的核心问题是智能决策问题,智能决策的基础是作战计划的推演和行动计划的优化。演化博弈强化学习模型通过引入演化博弈,使模型能够充分考虑对手的博弈意志和指挥艺术,并且在求解过程中可以使用现有的强化学习算法进一步提升模型的效率。针对离散和连续两种不同场景,分别推导出对应的复因子动力学方程,并给出了非对称博弈问题的一般性求解策略,摆脱了“理性人”这一假设的演化博弈强化模型。相较于现有模型,表征精度更好,刻画博弈对抗更精确,更符合军事对抗实际,有助于智能蓝军的精准化建设,相关结论在多智能体建模平台NetLogo上进行了验证。

The core of intelligentized command and control is intelligent decision making.The basis of intelligent decision making is the deduction of the combat plan and the optimization of the action plan.By introducing the evolutionary game,the reinforcement learning model of evolutionary game makes itself fully consider the gaming will and commanding art of the opponents,and able to use the existing reinforcement learning algorithms in the solution process to further improve the efficiency of the model.For discrete and continuous scenarios,the corresponding complex factor dynamics equations are derived respectively,and the general solution strategies for asymmetric game problems are given.Compared with the existing models,the enhanced model of evolutionary game without rational man assumption has better representation accuracy,more accurate depiction of game confrontation,and more in line with the reality of military confrontation,which will help the precision construction of the intelligent Blue Army.The relevant conclusions have been verified on the multi-agent modeling platform NetLogo.