Privacy-Preserving Load-Shaping Strategies for Smart Meters using Deep Reinforcement Learning

Smart meters improve household energy management but raise privacy concerns by revealing detailed consumption patterns. Previous methods modify smart meter readings but fail to maintain consistent energy levels in storage devices after protection cycles. In this work, we propose a novel proactive load-shaping strategy based on deep deterministic policy gradient (PLS-DDPG) designed to enhance users’ privacy by creating artificial load patterns while managing energy efficiently. Our method uses a novel reward-shaping mechanism to incentivize privacy-centric behaviors and introduces a battery consistency mechanism. Evaluated against a state-of-the-art adversary algorithm targeting appliance disaggregation, our approach performs on par with previous methods in privacy protection. However, our proposed approach retains 96.55% of the electricity capacity compared to 67.24% achieved by the state-of-the-art work, highlighting its superior efficiency in managing energy resources while upholding privacy.