A solution for multi-agent resource sharing based on fuzzy logic and particle swarm optimization

Abstract

Learning in multi-agent settings is a hard task, since agents learn simultaneously and the actions of each agent affect the environment of the others (and vice versa). In real world applications, such as multi-agent settings that deal with natural resource sharing, additional constraints are imposed that radically increase the complexity of learning. In this paper, we propose a solution for multi-agent resource sharing using a novel Fuzzy Inference System (FIS). The proposed FIS is optimized using Particle Swarm Optimization, that estimates the membership functions of the fuzzy variables. Experiments were conducted using a Monte Carlo procedure, comparing the performance of typical multi-agent policies with and without the proposed FIS, when applied on a real world resource sharing multi-agent model. Results demonstrate that the proposed FIS is capable of revealing efficient resource allocation schemes, that preserve the resource and, at the same time, ensure the benefit of the agents community.