Self-adaptation of multi-agent systems in dynamic environments based on experience exchanges

In complex and changing environments, MASs (Multi-Agent Systems) have to be faced with many challenges and their adaptivity can hardly emerge from the localized, phased and timely behaviors and decisions of agents. In this paper, a self-adaptation mechanism based on experience exchanges is proposed for MASs. In the mechanism, agents can dynamically induce environmental constraints and behavior patterns for generating new adaptation policies from two types of experiences exchanged by others, i.e., immediate and retrospective experiences, and then build action matrix based on the adaptation policies for determining their behaviors that can positively contribute to the improvement of global adaptivity of MASs. In the end, an example MAS system simulating a delivery system is described and experiments are conducted on the system to validate the mechanism for self-adaptive MASs.