Predicting the Power Consumption of Electric Appliances through Time Series Pattern Matching

We present a system to forecast the power consumption of electric household appliances. Accurate load prediction has numerous application domains, e.g., the facilitation of peak load prediction at a much higher resolution than permitted by state-of-the-art load profiles. Our solution is based on the identification and isolation of representative characteristic signatures from previously collected power consumption traces. Subsequently, time series pattern matching is applied to detect these signatures in real-time data, and emit predictions of an appliance's future consumption based thereupon. We evaluate the prediction accuracy of our approach with thousands of device-level power consumption traces and highlight the achievable prediction horizon.