供应链的快速发展带动了供应链采购电子化、数字化的需求,处于主导地位的供应链节点企业开始寻求更高效率的采购谈判方式以应对高频率的采购需求,而现有模型难以完全支持高复杂度的实时谈判场景.本文面向供应链多边采购,对自动谈判模型进行探索.首先,对多属性一对多自动谈判模型进行模块拆分,在对对手信息进行预处理的基础上,采用连续变量的贝叶斯方法估计谈判对手的偏好,促进多边多属性谈判达成共赢局势.其次,在策略生成部分引入 TD3(twin delayed deep deterministic policy gradient algorithm,双延迟深度确定性策略梯度)强化学习算法,实现了 agent对谈判过程的感知和对谈判对手的策略行为判断.然后,引入增强拉格朗日算法输出能够实现谈判双方联合效益最大化的反馈提案.最后,本文通过人机实验验证了模型在多场景下的谈判策略可行性和有效性.
An automated negotiation model for multilateral multi-issue procurement
The rapid development of the supply chain has driven the demand for the electronic and digital procurement of the supply chain.The leading supply chain node enterprises have begun to seek more efficient negotiation measures to cope with the high frequency of procure-ment demand.In contrast,the existing automatic negotiation models cannot fully support the increased complexity of real-time negotiation scenarios.This paper explores the automated nego-tiation model for multilateral procurement in the supply chain.First,the multi-issue one-to-many automated negotiation model is divided into modules.Based on the pretreated opponent infor-mation,the Bayesian method of continuous variables estimates the negotiators'preferences and promotes the multilateral multi-attribute negotiation to reach a win-win situation.Secondly,in the strategy generation part,twin delayed deep deterministic policy gradient algorithm(TD3)is introduced to realize the agent's perception of the negotiation process and the judgment of the negotiator's strategic behavior.Then,the enhanced Lagrange algorithm is introduced to output the counter-proposal that can maximize the joint benefits of both parties.Finally,this paper ver-ifies the feasibility and effectiveness of the negotiation strategy model through human-computer experiments in various scenarios.
NETL
NSTL
万方数据
