摘要
针对博弈强化学习中环境、信息、和激励函数的不确定性问题,通过对现有博弈强化学习算法的仔细研究和横向比较,以确定性方案、即时方案和适度推理方案3个角度对算法和模型进行系统归纳梳理,剖析多学科领域知识是如何相互融合并解决博弈强化学习的各类不确定性问题,指出博弈强化学习研究的重难点和今后的重点发展方向。结合模糊推理系统和分形与分数阶微积分理论给出一些新型解决思路。
Aiming at the uncertainty of the environment,information,and incentive function in game reinforcement learning,through careful research and horizontal comparison of existing game reinforcement learning algorithms,the three perspectives of the deterministic scheme,the immediate scheme and the moderate reasoning scheme were used to evaluate the algorithms and the models.The results of the research analyze how multi-disciplinary domain knowledge integrates with each other to solve various uncertainties in game reinforcement learning,and point out the difficulties and future key development directions of game reinforcement learning research.Some new solutions were given by combining the fuzzy reasoning system and fractional calculus theory.
作者
陈英
王军
陈希亮
张启阳
CHEN Ying;WANG Jun;CHEN Xi-liang;ZHANG Qi-yang(Postdoctoral Research Workstation,Eastern Theater General Hospital,Nanjing 210002,China;School of Software Engineering,Jinling Institute of Technology,Nanjing 211169,China;College of Command Information System,Army Engineering University,Nanjing 210001,China)
出处
《计算机工程与设计》
北大核心
2023年第11期3477-3488,共12页
Computer Engineering and Design
基金
金陵科技学院高层次人才基金项目(jit-b-201708)
国家自然科学基金项目(61806221)
国防科技重点实验室基金项目(6142101180304)。
关键词
强化学习
博弈论
不确定性
纯策略纳什均衡
分形
模糊系统
智能决策
reinforcement learning
game theory
uncertainty
pure strategy Nash equilibrium
fractal
fuzzy system
intelligent decision
