Enhanced spatio-temporal electric load forecasts using less data with active deep learning

Arsam Aryandoust, Anthony Patt, Stefan Pfenninger

Abstract: An effective way to mitigate climate change is to electrify most of our energy demand and supply the necessary electricity from renewable wind and solar power plants. Spatio-temporal predictions of electric load become increasingly important for planning this transition, while deep learning prediction models provide increasingly accurate predictions for it. The data that are used for training deep learning models, however, are usually collected at random using a passive learning approach. This naturally results in a large demand for data and associated costs for sensors such as smart meters, posing a large barrier for electric utilities when decarbonizing their grids. Here we investigate whether electric utilities can use active learning to collect a more informative subset of data by leveraging additional computation for better distributing smart meters. We predict ground-truth electric load profiles for single buildings using only remotely sensed data from aerial imagery of these buildings and meteorological conditions in the area of these buildings at different times. We find that active learning can enable 26–81% more accurate predictions using 29–46% less data at the price of 4–11 times more computation compared with passive learning. To enable electrification of cities, thereby decarbonizing energy grids, it is essential to predict electricity consumption with high spatio-temporal accuracy. To reduce the need for large amounts of training data, an active deep learning approach is developed to make accurate forecasts of electric load profiles at the scale of single buildings.