Studying Utility Metrics for Differentially Private Low-Voltage Grid Monitoring

The low-voltage energy distribution grid carries power to industrial and residential customers. To ensure its correct operation, distribution grid operators aim to monitor the grid with grid monitoring systems continuously. The system processes the stream of active and reactive power measurements at customer connections. Since this imposes major privacy concerns, data gateways typically sanitize the streams by adding noise to each measurement to achieve differential privacy. This however reduces the utility of the grid monitoring system. Due to missing studies, the utility of grid monitoring is not known. This leads to two research questions investigated in this study by means of a realistic case study. The first question is how to measure the utility appropriately. The second question is to give an intuition on whether one can achieve reasonable privacy and utility at the same time. Studying these questions is challenging for two reasons: The plurality of (1) grid analyses a grid monitoring system conducts, and (2) privacy requirements customers can have. To tackle the challenges, we identify a set of candidate utility metrics and use a differential privacy mechanism that unpacks multiple privacy requirements into one scaling parameter. Our experiments on a real-world grid and realistic measurements indicate the following. First, the utility of grid monitoring decreases faster than the sanitization error, that is frequently used in related work on differential privacy as utility metric. Second, already under weak privacy requirements, the utility is lower than under measurement errors.