Abstract:

To solve the problems of high global sensitivity, complex calculation of query on electricity consumption data and that independent noise can be easily filtered, Period Sensitivity Method（PSM） is proposed. PSM decomposes the consumption series into stationary series set and active series set according to the data statistical characteristics. According to the data sparsity and correlation, two disparate privacy protection strategies are put forward. It adds the independent identically distributed noise to the stationary series, then the stationary series with noise passes through the smoothing filter to obtain better utility. It adds the correlated noise to the active series?to resist the filtering attack, whose auto-correlation function is similar to active consumption series. Theoretical analysis and experimental evaluation show that PSM satisfies differential privacy and has better utility and smaller computational complexity.

Key words: correlated differential privacy, privacy release, smart meter, period sensitivity, series decomposition

摘要：

针对查询函数在用户用电数据上的全局敏感度较大、计算复杂度较高且独立噪声易被滤除的问题，提出了一种基于周期敏感度的差分隐私保护方法（Period Sensitivity Method，PSM）。PSM将用电序列分解为稳定期序列集和活跃期序列集，并根据数据稀疏度和相关性的差异使用两种隐私保护策略。向稳定期序列添加独立同分布噪声，并使用平滑滤波器对加噪后的稳定期序列进行平滑处理；向活跃期序列添加与活跃期序列的自相关函数相同的相关性噪声。理论分析与实验结果表明，PSM满足差分隐私，并且具有更好的可用性和更小的计算复杂度。

关键词: 相关性差分隐私, 隐私发布, 智能电表, 周期敏感度, 序列分解