Large language models (LLMs) have the potential to significantly enhance building data exchange. They hold special promise for textual data like energy efficiency measures (EEMs), but have not yet been applied in this domain. To address this gap, a novel methodology was developed using LLMs to parse and compare two distinct EEM lists. EEM names in both lists were processed through an LLM, and, for each EEM in the first list, the most similar EEM in the second list was identified. The results showed considerable alignment between the model's top-predicted matches and the best match identified manually, demonstrating the value of LLMs for building data exchange.

It is challenging to coordinate multiple distributed energy resources in a single or multiple buildings to ensure efficient and flexible operation. Advanced control algorithms such as model predictive control and reinforcement learning control provide solutions to this problem by effectively managing a distribution of distributed energy resource control tasks while adapting to unique building characteristics, and cooperating towards improving multi-objective key performance indicator. Yet, a research gap for advanced control adoption is the ability to benchmark algorithm performance. CityLearn addresses this gap an open-source Gym environment for the easy implementation and benchmarking of simple rule-based control and advanced algorithms that has an advantage of modeling simplicity, multi-agent control, district-level objectives, and control resiliency assessment. Here we demonstrate the functionalities of CityLearn using 17 different building control problems that have varying complexity with respect to the number of controllable distributed energy resources in buildings, the simplicity of the control algorithm, the control objective, and district size.

Increasing energy consumption has led to wide-scale optimization efforts of urban and sub-urban environments. To supplement this, building-energy-modeling (BEM) methods are applied to provide insights and identify optimizations. Several inputs are required for BEM, including climate, usage, and most importantly geometric data. Creation of geometric data is a time-consuming endeavor. Methods of automation often incorporate expensive or not widely available technology such as light detection and ranging (LiDAR) or unmanned aerial vehicle (UAV) imagery. This document aims to provide and prove the viability of a methodology to create building models at high levels-of-detail (LOD), using only easily available sources such as OpenStreetMaps (OSM) and street-view images (SVIs). Modern image processing techniques and machine learning algorithms such as convolutional-neural-networks (CNNs) and regional-convolutional-neural networks (R-CNNs) are explored with the goal of creating building geometry model for energy modeling with machine learning algorithms reaching a precision of 71\%, with geometry creation reaching an average accuracy of 95\%. The process can easily be extended to be user-assisted to greatly increase overall accuracy and reduce time complexity of the workflow for building geometry creation, outputting point-cloud data which will be ultimately applied for BEM.

The rapid progression in artificial intelligence has facilitated the emergence of Large Language Models (LLMs) like ChatGPT, offering potential applications extending into building energy modeling (BEM). This paper investigates the innovative integration of LLMs with BEM tools, focusing specifically on the fusion of ChatGPT with EnergyPlus. A literature review reveals a growing trend of incorporating LLMs in engineering modeling, albeit limited research on their application in BEM. We underscore the potential of LLMs in addressing BEM's challenges and outline potential applications such as input generation. Through case studies, we demonstrate the transformative potential of LLMs in revolutionizing the BEM lifecycle.

This paper presents a novel approach for developing irregular shading system. The Reinforcement Learning (RL) based approach balances the indoor light levels across multiple indoor locations and surface planes. The development of irregular shading system focuses on real-world scenarios where indoor illuminance value changes hourly throughout the day, and the shading system must respond in real-time. The proposed global simulation framework integrates building geometry, lighting simulation, and our RL model, and the learning process is achieved by a closed-loop training process. The results demonstrate that our approach is reliable in guiding different dynamic facade designs.

Early-stage optimization is effective in reducing a building’s energy consumption. However, today’s optimization process is based on simulation and is often very time-consuming and inefficient. To address this, we developed a machine learning (ML) surrogate model to replace the heating demand simulation process. The model was trained using a parametrically generated synthetic dataset based on rectangular buildings. We investigated which ML algorithm performs best concerning the size of the training dataset. We found that Linear Regression performs best when the dataset is smaller than 1000 while Random Forest performs best as the training dataset increases above 1000. When the dataset size reaches 9200, Random Forest can reach a mean absolute error of 1.68 kWh/m2 and a root mean square error of 2.69 kWh/m2. The best-performing surrogate model can predict the heating demand within 0.00005 seconds with high accuracy. Thus, our results show that an ML surrogate model can substitute a building’s heating demand simulation and significantly improve the efficiency of early-stage optimization processes.

We present a modeler-assistant tool designed to facilitate the creation and validation of building and district energy models. Our objective is not to develop a new modeling library but rather to enhance the usability of open-source modeling libraries and the efficiency of the model creation and simulation process. This methodology significantly reduces the manual effort required in model validation. It leverages Large Language Models (LLMs) and Python programming to transform textual description of requirements into Modelica models, complete with graphical annotations for thorough inspection. A key innovation is our API, which leverages syntactically constrained LLMs to streamline user interaction with the Python-based Modelica model generator, ensuring the consistent creation of valid models. Our approach has the potential to reduce manual effort for model validation from hours to minutes, significantly streamlining the process. We provide examples of model generation at various levels of input detail and showcase the integration of weather and operational data to calibrate the models, aligning it more accurately with real-world scenarios.

Insufficient building information, including footprint, conditioned area, age, and type, hinders urban-scale energy modeling. These parameters are crucial inputs for the simulation and optimization processes integral to the modeling. Prototypical building energy models, based on building surveys and code requirements at the time of construction, are frequently used when audit-quality data is unavailable. This helps to infer internal building characteristics. However, even local data sources like tax assessors' data contain unique land use or parcel codes that can be challenging to map to these prototypical buildings. This information does not directly correlate with the standard building type used to perform energy simulations. In this study, we apply and cross-validate several machine learning algorithms to automate the mapping from general building descriptions to standardized building types, as defined by the U.S. Department of Energy (DOE), a key component to accurately estimate building energy profiles at scale. The XGBoost algorithm outperformed others, achieving an F1 score, precision, and recall of 92.8\%, 93.4\%, and 93.0\%, respectively. These results highlight the potential of advanced machine learning techniques in bridging the data gap for urban-scale energy modeling and suggest a path forward for enhancing the resolution and accuracy of large building energy datasets.