Home appliance detection from aggregated energy consumption data on a single circuit

Home appliance detection helps identify individual electrical and electronic appliances, which in turn can be used to foster more efficient energy usage habits and promote energy conservation, and enable load forecasting and demand response support. Prior approaches, however, require sub-metering high resolution power signals of individual appliances during training or testing, which is not practical as in practice such high resolution power signals would not be made available economically. We propose an alternative, low cost approach for home appliance detection that can be used to detect home appliances from the aggregated energy consumption data on a single circuit. In the first step, our proposed approach utilizes pattern matching algorithms to recover power from the minute level time series of aggregated energy use metered on a single circuit. In the second step, a domain knowledge based approach is proposed to label individual appliances based on matching with the known signatures for the appliances. Our proposed approach is validated using a dataset from Clarkson University's Smart Housing Project, where we apply the proposed approach to the dataset and measure precision and recall. The evaluation shows that our approach is promising in detecting those appliances that contribute the predominate energy consumption, such as mini-fridges, desktops, lamps, and laptops. We also analyze in detail the reasons why some appliances are harder to detect correctly.