Power Baseline Modeling for Split-Type Air Conditioner in Building Energy Management Systems Using Deep Learning

The baseline model, an estimation of power or energy consumption before implementing energy management, is widely used to identify savings by comparing with the measured data after implementing energy management. This paper proposes power baseline modeling for split-type air conditioners using deep learning. Raw data is cleaned and prepared using data pre-processing. Two feature selection techniques, Karl Pearson's coefficient of correlation and Random forest, are used to select appropriate variables as inputs for the baseline modeling. Deep learning has been applied to modeling and estimating the split-type air conditioner power baseline. Performances of the deep learning model is evaluated using Coefficient of Variation of the Root Mean Square Error (CV-RMSE), Normalize Mean Bias Error (NMBE) and R-squared (R²). The results show that the performance indices are in acceptable accuracy according to the baseline modeling criteria.