Loading...
Multi-Objective Task Allocation Via Multi-Agent Coalition Formation
Amer, Noha Tarek
Amer, Noha Tarek
Description
A Master of Science thesis in Engineering Systems Management by Noha Tarek Amer entitled, "Multi-Objective Task Allocation Via Multi-Agent Coalition Formation," submitted in May 2012. Thesis advisor is Dr. Fouad Ben Abdelaziz. Available are both soft and hard copies of the thesis.
Abstract
Nowadays, tasks are complex and cannot be performed by individual agents. Therefore, there is a need to form coalitions utilizing the resources in order to maximize the efficiency of the system and/or maximize the payoff of each agent. In this thesis, we will formulate an optimization model to form optimal coalitions of agents satisfying both maximizing the efficiency of the system for cooperative settings and/or the payoff of the agent for non-cooperative settings. For small sized problems, we propose an optimization model to get exact solutions. For large sized problems, we propose genetic algorithms to get satisfying solutions. Different experiments were performed for the different settings. For cooperative settings, we observe that the tasks with the maximum payment are being performed, the agents who are most capable are assigned to perform the task provided they are the cheapest, and agents are assigned if and only if they will participate in performing a task to avoid any wasted costs. As for the genetic algorithms, the code is giving exact solution for medium problems. For selfish settings, we observe that the tasks with the maximum payments are being performed because they give maximum payoffs for agents, and agents prefer high payment over a complex task; that is, a capable agent would perform a high paying complex task rather than a low paying simple task. Also, very frequently all agents are assigned to tasks even if they don't contribute to accomplishing the task just to increase their payoff. As for the genetic algorithms, the code is giving exact solution for medium problems with time savings advantages. As for the hybrid setting, we observe that by comparing both cooperative and selfish agents, cooperative agents provide better results for the system. Our results coincide with the theories of game theory that say cooperative games provide a higher utility. However, we also show that combining both behaviors of agents, cooperative and selfish, provides better results. In addition to that, our genetic algorithm is giving exact solution for medium problems with time savings towards problems with a large number of agents. Most applications are realized in e-commerce systems or parallel processing. Search Terms: Multi Objective Programming, Coalition Formation, Task Allocation, Genetic Algorithms, Cooperative Agents, Non-Cooperative Agents