Finding near-optimal strategies for negotiation with incomplete information using a diversity controlling EDA

Negotiation activities are carried out among stakeholders in a supply chain for resolving the differences in their objectives and preferences. The objective of this work is to find near-optimal negotiation strategies for bilateral negotiation with incomplete information, in which each agent has a competitive objective and incomplete information about its opponent, by coevolving both agents' strategies using an estimation of distribution algorithm (EDA). However, an EDA often cannot find optimum solutions for both the agents for the reason that one of the two agents' populations has a very rapid convergence rate than the other population in the process of coevolution. Hence, the resulting solutions of both the agents will be biased and converge to suboptimal ones. To solve this problem, this paper proposes an EDA which has a novel diversity controlling capability. The proposed method utilizes the accumulated frequency information of the occurrence of individuals in each band of a population for generations. The information is used in the diversification and refinement procedure of the proposed diversity controlling EDA. Results from a series of experiments indicate that the proposed diversity controlling EDA achieves a better performance than the EDA without a diversity controlling method for the problem task in finding near-optimal solutions.