Accelerating Urban Building Energy Modeling

Abstract

Enabling data-driven decision-making in the built environment is critical to achieving ambitious and urgent decarbonization goals. In the building sector, urban building energy models (UBEMs) have become a valuable tool for jurisdictions to develop evidence-based retrofitting policies, but dynamically exploring solutions is hampered by the computational expense and organizational overhead of physics-based building energy models. In order to address these challenges, we present a fast, flexible, and comprehensive UBEM methodology which can be used to reduce identified barriers to time-sensitive decision-making in building stock decarbonization spheres. The methodology combines the speed of current data-driven approaches with the flexibility of computationally intensive, but accurate, engineering models. Identifying machine learning methods as a viable approach, we implement convolutional neural networks (CNNs) which embed timeseries from hourly weather data and building schedules; the embeddings are then combined with static building characteristics and projected to monthly heating and cooling loads. The proposed approach allows for programmatic flexibility and robustness to unique hourly weather conditions globally, while contextual abstraction enables geometric independence. A dataset of over 1 million detailed thermodynamics-based simulations was constructed to train and validate the surrogate model. Model results at the individual shoebox, building, and urban scales compare favorably to traditional numerical methods and meet accepted error bounds under national energy simulation standards. Additional validation at the urban- and national-scales are performed using public building simulation datasets. We then demonstrate expanded applications, which leverage the reduced computational cost of the framework to make traditionally infeasible analysis modes tractable and deployable. The methodology presented is intended to be utilized for both very-large-scale systematic analysis and near-real-time interactive explorations. In developing this framework, we aim to provide new mechanisms for key stakeholders in the decarbonization effort to quickly generate actionable insights and engage in iterative discussions to develop evidence-based policy across global building stocks.