RL-BLH: Learning-Based Battery Control for Cost Savings and Privacy Preservation for Smart Meters

Abstract

An emerging solution to privacy issues in smart grids is battery-based load hiding (BLH) that uses a rechargeable battery to decouple the meter readings from user activities. However, existing BLH algorithms have two significant limitations: (1) Most of them focus on flattening high-frequency variation of usage profile only, thereby still revealing a low-frequency shape, (2) Otherwise, they assume to know a statistical model of usage pattern. To overcome these limitations, we propose a new BLH algorithm, named RL-BLH. The RL-BLH hides both low-frequency and high-frequency usage patterns by shaping the meter readings to rectangular pulses. The RL-BLH learns a decision policy for choosing pulse magnitudes on the fly without prior knowledge of usage pattern. The decision policy is designed to charge and discharge the battery in the optimal way to maximize cost savings. We also provide heuristics to shorten learning time and improve cost savings.