基于深度学习的动态数据增量式挖掘方法

Incremental mining method of dynamic data based on deep learning

为了降低数据挖掘时间,提升数据挖掘精度,提出基于深度学习的动态数据增量式挖掘方法。利用简单处理单元组合构成并行分布式处理器,采用栈式稀疏降噪自编码网络作为动态数据特征提取的深度学习模型,通过逐层贪婪无监督策略完成预训练,使用随机梯度下降优化网络权值,获得动态数据规律。利用粗糙集求解近似动态约简容错偏差与置信区间,对动态数据样本子集进行增量式约简计算,建立区分矩阵与逻辑解析式,通过计算得到合取范式,把核属性引入各合取项中,实现动态数据增量式挖掘。仿真结果表明,所提方法具有较高的挖掘精度与效率。

In order to reduce data mining time and improve data mining accuracy,a dynamic data incremental mining method based on Deep Learning is proposed.A combination of simple processing units is used to form a parallel distributed processor,and a stack-type sparse noise reduction auto-encoding network is used as a Deep Learning model for dynamic data feature extraction to complete the pre-training through a layer-by-layer greedy unsupervised strategy.Then,stochastic gradient descent is used to optimize network weights to obtain dynamic data laws.Besides,rough set is adopted to solve approximate dynamic reduction error tolerance deviation and confidence interval,and incremental reduction calculation is performed on a subset of dynamic data samples.In addition,discernible matrix and logical analytical formula are established to obtain conjunctive normal form through calculation,and combined with the introduction of core attributes into each conjunct item,which finally realizes incremental mining of dynamic data.The simulation results show that the proposed method has high mining accuracy and efficiency.