IN-24-C028 – Eplus-LLM: A Novel Automated Building Simulation Platform Using Natural Language

Building Performance Simulation (BPS) through Building Energy Models (BEMs) serves as a critical tool for various applications aimed at enhancing building energy efficiency and sustainability. However, the establishment of BEMs often proves to be labor-intensive and time-consuming due to requisite expertise in building science, building equipment, as well as software usage. To address these challenges and facilitate user-friendly human-machine interaction for automated building simulation, we introduce Eplus-LLM (EnergyPlus-Large Language Model). This innovative approach is built upon a fine-tuned large language model (LLM) designed to assist modelers in building design and simulation tasks. Leveraging the attention mechanism within the LLM, Eplus-LLM enables modelers to engage in natural language interaction, allowing the model to comprehend the modeler’s demands and map human language into precise simulation models. By using scripts to call simulation software application programming interfaces (APIs), i.e., using Python to invoke the EnergyPlus engine without the need to utilize EnergyPlus for simulation, Eplus-LLM automates the creation of building models and generates simulation results efficiently. Validation results demonstrate that our proposed Eplus-LLM can generate BEMs encompassing different geometries and various internal load settings. The generated model structure and simulation results align seamlessly with experts’ modeling, affirming the effectiveness and robustness of our approach. In this study, we customize the LLM for the purpose of automated building modeling for the first time, directly constructing building models from natural language. This approach significantly improves the accessibility of building simulation, offering modelers a simple and efficient means to interact with the simulation software and obtain building models and simulation results. Moreover, our research has the potential to serve as a prototype for applications in other fields, with the prospect of being disseminated publicly or implemented in practical business settings.