Under the background of pricing in electronic markets, a multi-agent reinforcement learning algorithm based on fuzzy inference is proposed. Within Markov stochastic game framework, domain knowledge is initialized into fuzzy rules. Agents choose their actions according to those rules, which are updated by reinforcement learning. By doing so, Domain knowledge is effectively integrated; each domain sample is effectively exploited; more importantly, the learning dimension is greatly reduced. Compassion with former pricing algorithm indicates that FI-MARL improves average pricing profits, both individually and collectively; agents acquire long-term intelligence around either the cooperation or the competition issue.%以电子市场智能定价问题为研究背景,提出基于模糊推理的多智能体强化学习算法(FI-MARL).在马尔科夫博弈学习框架下,将领域知识初始化为一个模糊规则集合,智能体基于模糊规则选择动作,并采用强化学习来强化模糊规则.该方法有效融合应用背景的领域知识,充分利用样本信息并降低学习空间维数,从而增强在线学习性能.在电子市场定价的对比实验中,智能体无论在合作还是在竞争的问题上都表现出较为长远的智能行为,提高了平均定价收益.
展开▼
