A Fast Approach to Form Core-Stable Coalitions Based on a Dynamic Model

Forming collectives of agents that operate together as a group is a crucial challenge for Multi-Agent Systems. In this paper we focus on a dynamic model for forming stable coalitions of selfish agents. Specifically, we consider the dynamic coalition formation approach proposed by Arnold and Schwalbe, where agents decides at each time step which of the existing coalition to join so to maximise their pay-off. We propose several key extensions to the approach to speed-up convergence and alleviate the computational burden for the agents. Our empirical analysis comprises two complementary scenarios: collective energy purchasing, where a set of energy consumer forms coalition to buy energy at reduced tariffs, and a synthetic benchmarking scenario. Our results in these scenarios indicate that our proposed extensions provide a speed-up in run time of up to four orders of magnitude with respect to the original approach.