动态数据挖掘的构造性机器学习方法研究

项目名称： 动态数据挖掘的构造性机器学习方法研究

项目编号： No.61273302

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 王伦文

作者单位： 中国人民解放军电子工程学院

项目金额： 80万元

中文摘要： 海量数据流的挖掘越来越受到相关部门的重视，由于噪声、干扰等因素的影响，挖掘难度越来越大，知识发现的准确性和时效性有待提高。本项目研究了基于构造型神经网络的数据流动态挖掘方法，实现对海量复杂数据动态挖掘。研究内容包括海量复杂数据动态挖掘算法总体框架，模糊相容商结构以及大数据集粒度划分方法，覆盖的最优划分，构造型神经网络全局优化算法，基于滑动窗口技术的数据流动态聚类和分类技术，数据流动态挖掘策略与过程规律提取方法研究。以宽频段无线电频谱监测数据挖掘为例，采用以上方法，动态选择样本、动态调整粒度和动态发现规则，快速发现异动信号等知识。本项目将以上不同方法有机结合起来解决大规模、复杂数据动态挖掘，为数据流挖掘提供一种结构变换的动态挖掘方法，可望为大规模、复杂数据流的动态挖掘难题提供解决途径。

中文关键词： 构造型神经网络；模糊相容商空间；数据挖掘；数据流；

英文摘要： Mining mass data stream plays an important role in many fields. Due to the influence of noise, interference and so on, data mining is more and more difficult. The accuracy and efficiency of data mining are expected to improve. In this project, dynamic data stream mining method based on constructive neural networks and fuzzy tolerance quotient space is put forward to mine mass and complex data stream. The main content are listed as follows: the dynamic mining algorithm framework of mass and complex data stream, fuzzy tolerance quotient space and granularity analysis of mass data, best partition for covering, global optimization algorithm of constructive neural networks, dynamic clustering and classification of data stream base on sliding window model, strategy of dynamic data stream mining and extraction method of procedural rule. Here in taking wide-band radio spectrum data mining for example, we dynamiclly choose training samples, adjust analytic granularity, find rules, and then quickly discover knowledge. The above approaches in this project are integrated to form a structural dynamic transformation method, which is expected to provide solution that can mine mass complex data stream efficiently.

英文关键词： constructive neural networks；fuzzy tolerance quotient space；data mining；data stream；