| Title |
Enhancing Building Energy Baseline Reliability Using Multiple Regression Change-Point Models with Principal Component Analysis |
| Authors |
송수원(Song, Suwon) ; 신지현(Shin, JiHyun) |
| DOI |
https://doi.org/10.5659/JAIK.2026.42.8.293 |
| Keywords |
Building Energy; Energy Baseline; Change-point Model; Multiple Regression Analysis; Principal Component Analysis |
| Abstract |
Linear regression change-point models are widely used for building energy baseline development because of their strong relationship with
outdoor air temperature. However, the reliability of baseline models is often influenced by additional meteorological variables, including
relative humidity and solar radiation. This study proposes a multiple regression change-point model incorporating outdoor air temperature,
relative humidity, and solar radiation, and compares its performance with a simple linear regression model, a linear regression change-point
model, and a principal component (PC)-based multiple regression change-point model using measured data from a case-study building. The
results indicate that the linear regression change-point and multiple regression change-point models achieve CV(RMSE) values of 20.48% (R²
= 0.635) and 18.95% (R² = 0.688), respectively, demonstrating substantial improvements over the simple linear regression model with a
CV(RMSE) of 28.93% (R² = 0.272). Principal component analysis (PCA) applied to the independent variables identifies PC1 (weather-related
component) and PC2 (temperature-level component), which together explain 90.5% of the total meteorological variance while effectively
resolving multicollinearity between relative humidity and solar radiation. In addition, PC3, retained based on regression coefficient significance
testing, further improves model performance by reducing the overall CV(RMSE) from 21.04% to 18.95%. These findings demonstrate that
multiple regression change-point models combined with independent-variable PCA and coefficient significance testing provide a robust and
reliable framework for developing building energy baselines from measured data. The proposed approach enables more stable and interpretable
baseline modeling under varying weather conditions compared with conventional temperature-based baseline methods. |