성과공개

4차

구성기술3

경희대학교

학술발표

설계 적합성 평가를 위한 인공지능 학습모델 구축 방안

한국CDE학회 2024 동계학술대회 논문집

김인한

이세진

Saddiq Ur Rehman; Syed Haseeb Shah; 김태원; 김동영

'This research presents a comprehensive AI-based Office Recommendation System, developing five distinct models to enhance the design conformity assessment based on client needs. The three Natural Language Processing (NLP) models—Word2Vec (trained on GloVe Twitter data), BERT (Bidirectional Encoder Representations from Transformers), and GPT—are employed to analyze client text inputs, extracting essential parameters for office building requirements. These NLP models transform textual data into meaningful parameters, laying the foundation for a robust recommendation system.
The recommendation system employs similarity metrics, including cosine similarity and Euclidean similarity, to match extracted parameters with suitable designs from a Building Information Modeling (BIM) database. This ensures a personalized and precise recommendation tailored to client needs.
Complementing the NLP models, two generative algorithms—Generative Adversarial Network (GAN) for facade generation and Genetic Algorithm (GA) for parking space layout generation—provide creative design solutions based on client input.
This holistic approach, incorporating NLP for text analysis, recommendation systems for design suggestions, and generative algorithms for creative output, establishes an advanced AI-driven Office Recommendation System, revolutionizing the design conformity assessment process for client-oriented office buildings.