The potential of smart home sensors in forecasting household electricity demand

The aim of this paper is to quantify the impact of disaggregated electric power measurements on the accuracy of household demand forecasts. Demand forecasting on the household level is regarded as an essential mechanism for matching distributed power generation and demand in smart power grids. We use state-of-the-art forecasting tools, in particular support vector machines and neural networks, to evaluate the use of disaggregated smart home sensor data for household-level demand forecasting. Our investigation leverages high resolution data from 3 private households collected over 30 days. Our key results are as follows: First, by comparing the accuracy of the machine learning based forecasts with a persistence forecast we show that advanced forecasting methods already yield better forecasts, even when carried out on aggregated household consumption data that could be obtained from smart meters (1-7%). Second, our comparison of forecasts based on disaggregated data from smart home sensors with the persistence and smart meter benchmarks reveals further forecast improvements (4-33%). Third, our sensitivity analysis with respect to the time resolution of data shows that more data only improves forecasting accuracy up to a certain point. Thus, having more sensors appears to be more valuable than increasing the time resolution of measurements.