Abstract: The paper discusses the problem of seller’s intelligent pricing in B2B marketplaces. on the basis of myopic adjustments by best response made by each seller, it puts forward an practical online learning method based on episode training. The method integrates both immediate result by reinforcement learning and rationality by deliberate inference, thus improves its online performance. The algorithm was proven to be applicable by simulations.

摘要： 以B2B电子市场中卖方agent的智能定价问题为应用背景，在库诺特短视调整基础上，应用Q学习算法，提出了基于情节序列训练的学习方法，将纯粹以结果为反馈的强化学习方法和以推理为目标的慎思过程结合起来,提高了算法的在线学习性能。仿真实验验证了算法的有效性，为推向实际应用奠定了基础。