A multi-agent method for periodicity detection in distributed events

Abstract

Multi-agent systems working in constantly changing environments may be challenged by instantaneous unavailability of their autonomous agents caused e.g. by limited computing resources. A new method of self-organization of distributed service components is proposed, suitable for multi-agent systems. This method relies on particular agents carrying out separate analyzes of their individual processing loads or other specific events. Analysis is done separately by each agent, and is based only on information about agent's own history of involvements in former tasks' execution. Agents can also decide to replace peers should such need arise. The superposition of individual agents' readiness periods emerges as a collective forecast of service load. By individual agents' decisions appropriate amounts of agents get ready to execute forthcoming tasks.