모델기반 예측제어를 위한 Grey-box 모델링의 LLM 기법 적용 가능성 검토 연구

초록

This study investigates the applicability of large language model (LLM) assistance to control- oriented grey box thermal modeling for future model predictive control (MPC) applications. Three modeling methods are compared under identical measured data conditions comprising a hand-coded RC grey box, an AI-generated RC grey box, and an AI-generated ANN black box. The experiments used measured sensor data obtained at 15 min intervals from four building zones and considered the official L1-L4 information levels. The performance was evaluated using the root mean squared error (RMSE) and the coefficient of variation of the root mean square (CVRMSE) for both one-step prediction and 24-h recursive rollout prediction. The results showed that the performance of the AI-generated RC grey box was more consistent across zones and information levels than the AI-generated ANN black box. Meanwhile, the artificial neural network (ANN) remained competitive in some cases but showed larger variability. These findings indicate that LLM assistance can support structured and reproducible grey-box workflow development without the need to replace conventional engineering modeling.

키워드