SUMMARY

As intermittent renewable energy generation resources become more prevalent, innovative ways to manage the electric grid are sought. In the past, much of the grid balancing effort has been focused on the supply side or on demand-side management of large commercial or industrial electricity customers. Today, with the increase in enabling technologies such as Internet-connected appliances, home energy management systems, and advanced metering infrastructure (AMI) smart meters, residential demand-side management is also a possibility. For a utility to assess the potential capacity of residential demand-side flexibility, power data from controllable appliances from a large sample of houses is required. These data may be collected by installing time- and cost-intensive monitoring equipment at every site, or, alternatively, by disaggregating the signals communicated to the utility by AMI meters. In this study, non-intrusive load monitoring algorithms are used to disaggregate low-resolution real power signals from AMI smart meters. Disaggregation results using both supervised and unsupervised versions of a graph signal processing (GSP) -based algorithm are presented. The effects of varying key parameters in each GSP algorithm, including scaling factor, sequence, and classifier threshold are also presented, and limitations of the algorithm based on energy use patterns are discussed. FM values greater than 0.8 were achieved for the electric resistance water heater and electric vehicle charger using the unsupervised GSP algorithm. The disaggregated signals are then used to develop energy forecasting models for predicting the load of controllable appliances over a given demand response period. ARIMA, SVR, and LSTM forecasting methods were evaluated and compared to a baseline model developed using the mean hourly power draw values. The minimum MAAPE was achieved for the water heater, with an approximate range of 10 < MAAPE < 20. The total energy flexibility of each appliance and the associated uncertainty of the combined disaggregation and forecast are characterized to assess the feasibility of this approach for demand-side management applications. The framework presented in this study may be used to characterize the ability of signals to be disaggregated from a larger dataset of AMI data, based on the whole-house signal characteristics. This analysis can aid grid managers in assessing the viability of selected devices, such as the water heater, for demand response activities. Remote Link: https://teams.microsoft.com/l/meetup-join/19%3ameeting_NTc0YTU5NmEtNjZlMi00MWUwLWJhOGEtYmVlODY2YzRiZWJi%40thread.v2/0?context=%7b%22Tid%22%3a%22482198bb-ae7b-4b25-8b7a-6d7f32faa083%22%2c%22Oid%22%3a%2264e73ef9-b15a-4193-a208-137060edd24f%22%7d