성과공개

4차

구성기술3

경희대학교

학술발표

시뮬레이션에서 예측까지: 파라메트릭 도구인 Ladybug Tools와 Deep Learning을 기반으로 한 빌딩 성능 최적화 연구

한국CDE학회 2024 하계학술대회 논문집

양우붕

김인한

황경은

Recently, deep learning technology has revolutionized building performance optimization design. Compared to traditional rule-based methods like Ladybug Tools, deep learning offers significant advantages in simulating building performance. This study introduces a novel approach that converts Ladybug Tools simulation results into graph structures suitable for Graph Neural Networks (GNNs). Representing the simulation mesh as a graph allows GNNs to capture complex data relationships. Additionally, Transformer models are proposed to process heterogeneous graphs, using their self-attention mechanism to focus on node relationships. This method enhances the model's ability to handle complex building geometries and irregular meshes, providing a foundation for future building performance prediction and optimization.