差分隐私保护下面向海量用户的用电数据聚类分析

智能电表实现了对用户用电信息全方位实时收集,使得对用户用电行为精准聚类分析成为了可能,然而在分析过程中易泄露用户信息。为此,提出了一种差分隐私保护下用户用电数据聚类分析的方法,该方法运用两阶段隐私保护聚类技术解决精准分析与隐私保护不能并存的矛盾。两阶段聚类采用了分布式的思想,包括局部聚类和全局聚类两部分。局部聚类采用了隐私保护下的自适应K-means算法对智能电表采集的原始用户用电数据进行初次聚类;全局聚类设计一种新的基于密度和层次思想的聚类算法用于对初次聚类结果进行二次优化聚类。相关实验表明了该方法的有效性。

Differential Privacy Protection Based Clustering Analysis of Electricity Consumption Data for Massive Consumers

Smart meter achieves real-time and comprehensive collection of user''s electricity consumption information. It is possible to accurately do clustering analysis of the user''s electrical behavior by using such data. However, in the process, the user''s information can be leaked easily. So a method of differential privacy clustering analysis for electricity information is proposed. It utilizes two phase privacy protection clustering to solve the contradiction that privacy protection and accurate analysis cannot coexist. Two phase clustering adopting the distributed computation idea consists of local clustering and global clustering. The former uses differential privacy adaptive K-means algorithm to complete first cluster electricity consumption data collected by smart meter. In global clustering, a new clustering algorithm based on density and hierarchy is designed to optimize the results from local clustering. Relevant experiments show that this method is effective.