Forecasting the Future of Flexible Buildings

Rapid changes are occurring in the electric utility sector. These changes include increasing renewable energy penetration, aging utility infrastructure, building electrification, and electric vehicle adoption, along with advances in smart and connected building technology. The new version of the electric utility sector is driving increased interest in dispatching energy use in buildings as grid assets. Enabling buildings to be dispatched like more traditional grid assets (i.e., battery electric storage systems) requires a completely new set of modeling, forecasting, and control methods. One key step to enable buildings to flex load is developing accurate forecasts of energy use in buildings. We must also understand how buildings can respond to dispatch signals from the grid to shed or shift load.

The rapid adoption of building automation systems (BAS) in commercial buildings over the past several decades has led to large amounts of time series data documenting building performance. This data has traditionally been siloed in proprietary BAS systems making it difficult to aggregate and optimize buildings. Recent development in network technology has unlocked data from commercial buildings. These developments include the BACnet communication protocol and smart building gateways. Data from residential buildings has also come online during the past decade as smart thermostats and smart home devices rapidly penetrated the residential market. Smart connected buildings are now seen as a key technical component for achieving the decarbonization goals in the building sector. These buildings also play a crucial role in enabling an efficient and resilient electricity grid.

Edo smart building data gateway. Source: Edo

High-resolution, short-term energy use forecasts provided by connected buildings enable utilities to accurately predict electricity demand. These forecasts enable utilities to take better pre-emptive action against constraints that occur in the generation, transmission, and distribution sections of the grid.

Edo is developing new machine learning methods for generating high-resolution, accurate forecasts of whole-building electricity demand. These forecast methods consider key predictors, including weather, occupancy patterns, and system operations. The predictors are used to describe trends in overall electricity demand. Forecasts are generated using statistical modeling and machine learning methods, such as auto-regressive models, gradient-boosting decision trees, and multi-layer neural networks. These methods deliver accurate, data-driven forecasts. Edo’s forecasting models are trained and tested against historical data from over 2,000 buildings to ensure that the most accurate models and features are used.

Edo is developing forecasting methods that have two-fold benefits. They improve electric utility awareness of load patterns. Furthermore, they optimize building controls across many interconnected equipment systems to improve efficiency. Our building controls and machine learning teams are developing methods that can be deployed quickly at scale across Edo-connected buildings. The methods assess, characterize, and predict the behavior of buildings. Understanding future operations and performance of buildings enables our systems to make controls adjustments to meet performance goals at the building and the grid levels. Our unique capabilities allow for the deployment of load shifting and load shedding strategies to modulate the load profile of a building.

Baseline, efficiency, and flexible load curves. Source: Edo

Deploying these strategies in a single building has limited impact. However, Edo’s technology is designed to be deployed at scale. Edo’s tech stack – including our gateway device, equipment mapping tool, and forecasting strategies -is designed to enable quick deployment and rapid model training and tuning. By aggregating many forecasts, Edo supports utility needs across a distribution or local service. Furthermore, we support the building portfolio to engage in demand flexibility programs.

[1] Clayton Miller, Pandarasamy Arjunan, Anjukan Kathirgamanathan, ChunFu, Jonathan Roth, June Young Park, Chris Balbach, Krishnan Gowri, Zoltan Nagy, Anthony D.Fontanini & Jeff Haberl (2020) The ASHRAE Great Energy Predictor III competition: Overview and results, Science and Technology for the Built Environment, 26:10, 1427-1447, DOI:10.1080/23744731.2020.1795514