Research On Distributed Data Mining System Based On Multi-Agent

With the development of the Internet, the application of distributed system is widely deployed, and users focus on the effective information mined from these massive data in these distributed systems. The centralized data mining technology is difficult to effectively solve problems in distributed data mining, such as the distribution of data storage, data heterogeneous, large amount of data, network bandwidth small etc. For that distributed data mining system requires the following characteristics: 1) Distributed processing: It is not necessary for the system to centralize all distributed (in logic or in physics) storage data to the same data warehouse, and it is to reduce the occupation of the network bandwidth. 2) Parallel processing: the system should be able to mine all kinds of distributed data parallelly, instead of accessing different storage of data frequently. 3) Good compatibility: the system can be compatible with all types of data in the distributed systems. According to these characteristics, multi-agent system is an effective way used to solve distributed data mining with its flexible management mode, efficient parallel processing ability. The using of multi-agent system in distributed data mining has attracted more and more researchers'attention.This thesis addresses related theory of agent technologies and data mining, furthermore, researches the related achievements of multi-agent system and distributed data mining in recent years, then proposes a New Multi-Agent System for Distributed Data Mining (NMAS-DDM).In this model system, a layered management structure with agents is established to meet the requirements of distributed processing and compatibility in distributed data mining. Meanwhile, the system adopts a message transfer mode based on knowledge transmitting, in order to avoid the message occupies too much network bandwidth. This thesis makes further research in tasks assigning and scheduling, then puts forward the way of assigning tasks based on task quantity or dimension. On this basis, the NMAS-DDM formulated the message format to occupy the network bandwidth at minimum. Finally, this thesis improves ant colony algorithm as the mining algorithm of NMAS-DDM, and this algorithm further reflects the characteristics of flexible and efficient of Multi-Agent System while doing distributed data mining.