Building Electricity Load Forecasting via Stacking Ensemble Learning Method with Moving Horizon Optimization
The short-term forecasting of building electricity demand is certain to play a vital role in the future power grid. Given the deployment of intermittent renewable energy sources and the ever increasing consumption of electricity, the generation of accurate demand-side electricity forecasts will be valuable to both grid operators and building energy management systems. The literature is rich with forecasting models for individual buildings. However, an ongoing challenge is the development of a broadly applicable method for electricity forecasting across geographic locations, seasons, and use-types. This paper addresses the need for a generalizable approach to electricity demand forecasting through the formulation of a stacking ensemble learning method. Rather than using a single model to predict electricity demand, our method uses a weighted linear combination of forecasts from multiple sub-models. By learning the model weights in real-time using electricity demand data streams and a moving horizon training technique, the method is more robust than a single model approach. By applying our method to electricity demand data sets for 8 different buildings, we show that this data-driven approach is capable of producing accurate multivariate forecasts for building level applications.