The Price of Undersampling Nonintrusive Electrical Data

Field data from power system measurements is frequently disappointing. Utilities are geographically large (in some cases remote), interact with a wide range of users and use profiles, and contain a burgeoning diversity of electrical loads. Meters and data collection systems for applications such as billing and phasor measurement are tailored to gather information at a price point specific to their intended use. Information recorded from these monitors can be surprisingly difficult to interpret and repurpose for other utility applications. Many measurement systems internally record data at higher rates than the rate at which they save or log final data. This paper demonstrates the value of implementing intelligent preprocessing of power data that preserves information needed for a particular application. We compare “typical” recorded data to revised datasets to examine the effect of targeted data preprocessing on machine learning and load monitoring applications. Spectral envelope preprocessing and V-I trajectories are used with raw data early in the measurement process to permit a flexible trade-off between sample storage rate and resolution. The UK-DALE dataset and newly acquired data from ship microgrids provide case studies for this work.