Analysis of supervised graph signal processing-based load disaggregation for residential demand-side management

Abstract

With the widespread adoption of advanced metering infrastructure (AMI) smart meters in recent years, interest in using residential buildings as demand-side management resources has increased. To characterize the amount of energy available for demand-side management in residential buildings, non-intrusive load monitoring (NILM) may be applied to the low-resolution data collected by AMI smart meters to provide insight into the energy consumption of individual controllable appliance types within the house. The disaggregation of water heaters and HVAC system components is investigated in this study, as these appliances are the greatest consumers of energy in U.S. homes. A supervised graph signal processing (GSP) algorithm is applied to real power data for appliances sampled at a 15-minute rate, with results demonstrating that disaggregation sequencing, graph scaling factor, and graph classifier threshold values have a significant impact on supervised GSP NILM accuracy. Scaling factor (ρ) was varied from 20 to 200, graph signal classifier threshold value (q) was varied from 0 to 0.5, and six disaggregation sequences were considered. Varying these parameters resulted in a maximum range in accuracy, as measured by F-measure, of 0.0 to 1.0 for a single appliance, demonstrating that choosing a set of optimal parameters is critical for disaggregation accuracy.