An investigation into dynamic problem solving in a hybrid evolutionary market-based multi-agent system

Static resources allocation problems have been widely studied. More recently some of this attention has changed to focus on dynamic problems, where problem specifications, constraints or resources may change before a solution is obtained. This work examines an approach that combines a multi agent system based on a simulated market with evolutionary optimization. Previous work has showed the efficacy of such a hybrid approach, where the characteristics of agents are subject to evolutionary optimization. This work compares the multi agent only, and the hybrid system, when the problem is subject to random change during the attempt to find a solution. Results confirm the advantages of evolutionary optimization of agent rules in a static or dynamic environment, both in terms of tasks completed within a given time, and the cost per task completed. Surprisingly, an optimum amount of noise exists, that improves the performance of the multi agent or hybrid trading model.