Research On Dynamic Contract Net Protocol Based On The Ant Colony Algorithm

As an important department of the DAI (Distributed ArtificialIntelligence), MAS (Multi-Agent System) produced and developedrapidly in recent years. It is the key problem of the MAS that thecoordination, cooperation and negotiation among a group of agents tocomplete the task. The Contract Net Protocol, which proposed byRandall Devis and Reid G.Smith is a classical strategy about taskallocation. There are two roles: Manager and Contractor in the CNP.Usually, we use the strategy solve the problem about task allocation. Inreal application, the researcher proposed many kinds of expanding andimproving schemes based on basic CNP to avoid the excesscommunication load in the classical CNP. But the scheme invalidated orperformed not well enough, when the ability of individual agent changed.In order to adapt the dynamic environment better, we proposed MDCNP(a Dynamic Contract Net Protocol Based on the Multi-Pheromones).In the paper, we do some works for the MDCNP:1. We do lots of research about CNP, including some classicalimproved schemes and the open and dynamic attribution of the MAS to find the research subject. It is necessary to solve how to use CNP indynamic environment. So we do some improvement based on the DCNPin the paper.2. We use the model of response threshold of the AAS (Adaptive AntSystem) in the CNP, which is based on social property of this model, topropose MDCNP. And the experiment data shows the MDCNP isavailable.3. Further research, we find it is important to add a pool, with certainnumber capacity, to the agent in the MAS. It improved the capability ofthe system. And it is proved by the experiment.4. In the paper, we choose the Repast as the platform. And we do agreat deal of tests on the Repast. The results of the tests show theadvantages of the MDCNP about the execution time andcommunication's quantity and necessity to add the pool to agent.