Change point detection is important for many real-world applications. While sensor readings enable line outage identification, they bring privacy concerns by allowing an adversary to divulge sensitive information such as household occupancy and economic status. In this paper, to preserve privacy, we develop a decentralized randomizing scheme to ensure no direct exposure of each user's raw data. Brought by the randomizing scheme, the trade-off between privacy gain and degradation of change point detection performance is quantified via studying the differential privacy framework and the Kullback-Leibler divergence. Furthermore, we propose a novel statistic to mitigate the impact of randomness, making our detection procedure both privacy-preserving and have optimal performance. The results of comprehensive experiments show that our proposed framework can effectively find the outage with privacy guarantees.

翻译：变点检测在众多实际应用中具有重要意义。传感器读数虽能实现线路故障识别，却也带来隐私泄露风险——攻击者可能借此获取住户入住情况、经济状况等敏感信息。为保护隐私，本文开发了一种去中心化随机化方案，确保每个用户的原始数据不直接暴露。通过研究差分隐私框架与库尔贝克-莱布勒散度，量化了由随机化方案引起的隐私增益与变点检测性能退化之间的权衡。进一步地，我们提出了一种新型统计量以降低随机性的影响，使检测过程兼具隐私保护性与最优性能。综合实验结果表明，本文提出的框架能在保障隐私的前提下有效定位故障。