Infomorphism: Integrating A Forecasting Model for High-Resolution Building Energy Demand Prediction

The idea of integrating renewable energy into urban design, through advanced building technologies and architectural design approaches, has recently started to see more attention. Accurately predicting energy consumption at a higher time resolution provides opportunities to broaden the scope of problems that can be addressed in planning frameworks. Increased granularity in time resolution, however, brings attendant problems such as increased data volatility, increased data volume, and nonlinearity. In this work, these challenges are addressed through a machine learning model selection process, to best predict hourly building energy demand. The model is trained on five different building data sets. This work finds that, across all buildings tested, a Random Forest model provides the best predictive accuracy - up to a 78% improvement over comparable state-of-the-art models such as a dense deep learning network.