Employing Private Data in AMI Applications: Short Term Load Forecasting Using Differentially Private Aggregated Data

Paper in proceeding, 2016

Data collected by sensors in the advanced metering infrastructure (AMI) can be used for a multitude of applications offering better control and service of the electrical network. One major application is Short Term Load Forecasting (STLF) which helps in better managing the near future energy production based on predicted consumption. However, the fine-grained data collected can be used to infer sensitive information about the customers' lifestyle, which calls for attack-resistant Privacy Enhancing Technologies (PETs) to protect these data. In this study we propose a methodology which can be used to enhance an AMI application with the help of data that undergoes PETs and to evaluate the effect of this enhancement on the application's utility. We apply this methodology on the STLF application and we analyze the privacy concerns raised by the different types of information contributing to it. This is followed by an exploratory study focused on the effects of differentially-private aggregation on STLF methods. We show that the noise introduced, in the case of a bounded sensitivity, has a small effect on the forecast accuracy. These results are a step forward towards early adoption of PETs by energy companies and also by integration in lower-tier-SCADA systems that manage microgrids with energy production capabilities. This enables using the most of what AMI data has to offer while enhancing customers' privacy.