Computing Electricity Consumption Profiles from Household Smart Meter Data

In this paper, we investigate a critical problem in smart meter data mining: computing electricity consumption profiles. We present a simple, interpretable and practical profiling framework for residential consumers, which accounts for variations in electricity consumption at different times of day and at different external temperatures. Our approach is to isolate the effect of external temperature on electricity consumption and apply a time-series autoregressive model to the remaining signal. The proposed profiles may be used for making personalized energy-saving recommendations, detecting outliers, and generating very large realistic data sets for testing the scalability of smart meter data management systems. Using predictive power as a metric for the accuracy of consumption profiles, we show, using a real data set of 1000 homes, that our approach results in improved root-mean-squared prediction error compared to existing approaches.