Analysis and Demand Forecasting of Residential Energy Consumption at Multiple Time Scales

The Internet of Things (IoT) allows us to connect and monitor devices from virtually anywhere. Electric utility companies have been replacing the outdated analog meters with the new smart meter versions to automatically capture information about electricity consumption at a fine time granularity and transmitting it back to the utility provider. Energy demand forecasting is essential for Smart Grid operations. Ability to perform data analytics on the collected smart meter measurements and then predicting the electricity demands plays an important role in the utility companies' decision making for their system planning and operations. While fine granularity measurements could be useful for getting deeper insights into electricity usage patterns of different households, they might be not optimal for energy demand forecasting. We demonstrate the importance of considering and selecting different time scales in performing data analytics and demand forecasting of residential buildings. Using smart meters measurements collected from 114 residential apartments at 1 minute granularity over one year, and weather information for the same period, we design an automated process for building an efficient ensemble of linear regression models to forecast the future energy demands. This process identifies the linear portions of the daily usage patterns and creates the apartment clusters with similar usage profiles to optimize the forecasting accuracy of the designed linear regression models. It could be applied to different residential areas and geographical regions to produce the customized ensembles of fast and efficient linear regression models. Experimental results demonstrate that the proposed approach and designed performance models achieve good accuracy with 7%-19% of prediction error, and therefore could be used for optimizing the future energy distribution across the utility grid and making related critical pricing decisions.