基于博弈对抗复杂系统的决策建模与求解  

Decision Modeling and Solution Based on Game Adversarial Complex Systems

作　　者：姜嘉成 贾政轩 徐钊 林廷宇 赵芃芃 欧一鸣 Jiang Jiachen;Jia Zhengxuan;Xu Zhao;Lin Tingyu;Zhao Pengpeng;Ou Yiming(Beijing Simulation Center,Beijing 100854,China)

机构地区：[1]北京仿真中心,北京100854

出　　处：《系统仿真学报》2025年第1期66-78,共13页Journal of System Simulation

摘　　要：面对现代博弈越发呈现“大规模、高烈度、非全知、强博弈”的复杂态势,针对现如今的传统的博弈决策灵活性不足、迭代周期长的情况,以无人红蓝博弈为背景,实现无人博弈对抗复杂系统建模,并基于深度强化学习技术开展在以无人红蓝博弈为背景下的智能决策算法研究,借助深度神经网络以及Bellman最优性原理,实现对庞大解空间的高效搜索,构建复杂博弈场景下的最优智能决策,完成对决策智能体的网络结构和训练算法的设计,达到最优博弈效能,以实现策略演进与快速迭代,并通过仿真验证所提出算法的有效性、灵活性和泛化能力。In view of the complex situation of the current game which will be large-scale,high-intensity,not omniscient,and strong confrontation,and in response to the lack of flexibility and long iteration cycles in traditional game decision-making,the model of the unmanned complex game system is built according to the background of the unmanned red and blue game.Based on deep reinforcement learning technology,intelligent decision-making algorithms are studied in the background of unmanned red and blue games.With the help of deep neural networks and Bellman's optimal principle,the search of the huge solution space is more efficient,and the optimal intelligent decision is constructed in complex game scenes.The network structure and training algorithm of the decision-making agent are designed in order to achieve optimal game efficiency,strategy evolution as well as rapid iteration.And the effectiveness,flexibility and generalization ability of the proposed algorithm are verified through simulation.

关 键 词：无人博弈对抗 复杂系统建模 智能决策 深度强化学习 策略演进 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]