Fair Resource Allocation Using Multi-population Evolutionary Algorithm

Abstract

Resource allocation between selfish agents are performed under centralised and/or distributed mechanisms. However, there are issues in both cases. In centralised solution, although the resources are allocated in an efficient way, the allocation decisions may not be acceptable for some selfish agents making them reluctant to cooperation. In decentralised solution, although the problem is solved from each agent’s perspective, the allocation leads to an inefficient usage of provided resources. For example, such an issue is evident in a water network distribution system where different agents share the river water and a central planner (CP) maximises the social welfare to the whole system. Issue arises when the CP solution is not acceptable by some agents. Therefore, a mechanism should be devised to encourage each agent to accept the CP decision. This paper introduces a mechanism in re-distributing the CP revenue value amongst the competing agents based on their contribution to the CP value. To find each user’s contribution, this paper develops a parallel evolutionary search algorithm which enables the agents to autonomously solve their local optimisation problem whilst interacting with the other agents and the whole system. The search evolves towards a solution which is used as an incentive for calculating a fair revenue for each agent. The framework is applied to a river reach with five competitive users. Results show decentralised coupled centralised approaches has the potential to represent mechanisms for a fair resource allocation among competing self-interested agents.