The smart meters have been widely deployed to monitor customer usage profiles around the world, which can read the energy load of residents blue at the rate of per minute or per second. This fine-grained data would expose personal behaviors and other sensitive information to malicious adversaries. To address this privacy concern, battery-based load hiding (BLH) scheme blue was proposed and had been explored for several years. The differential privacy is a provable method to preserve the privacy against the adversaries of arbitrary computational power. At first, a battery-based intermittently differential privacy (IDP) scheme is proposed in this article and also prove that the IDP can achieve the differential privacy. Then, we develop another scheme — a Reinforcement Learning (RL) algorithm to guide the battery control policy to match the requirement of battery constraints and cost-saving in a better way. Finally, we integrate the IDP scheme with the RL algorithm into a complete RL-IDP scheme. The experimental results show that the privacy-preserving level performs well by our RL-IDP scheme while it can achieve cost-saving.

智能电表已被广泛部署用于监控全球客户的使用情况，可以以每分钟或每秒的速度读取蓝色居民的能源负荷。这种细粒度的数据会将个人行为和其他敏感信息暴露给恶意对手。为了解决这个隐私问题，提出了基于电池的负载隐藏（BLH）方案，并且已经探索了几年。差分隐私是一种可证明的方法，可以保护隐私免受任意计算能力的攻击。本文首先提出了一种基于电池的间歇差分隐私（IDP）方案，并证明了IDP可以实现差分隐私。然后，我们开发了另一种方案——强化学习 (RL) 算法，以指导电池控制策略以更好的方式匹配电池约束和节省成本的要求。最后，我们将 IDP 方案与 RL 算法集成为一个完整的 RL-IDP 方案。实验结果表明，我们的 RL-IDP 方案在隐私保护级别上表现良好，同时可以节省成本。