Convolutional Neural Networks for Predicting Energy Efficiency of Air Conditioning Systems

This paper presents a prediction of the energy efficiency of air conditioning system using convolutional neural network (CNN). Coefficient of Performance (COP) is used as a measure of energy efficiency. In previous work, the study employed autoregressive with exogenous inputs (ARX) as a model for COP prediction. In this paper, the obtained data are preprocessed, and COP is calculated and used in the training process. We employ CNN to build two sub-models to separately predict temperature difference and power consumption using the past 30-minute data of room temperature, ambient temperature, and power consumption. The models are tested with historical data and compute the root mean square error (RMSE) as a prediction indicator. We compare the proposed CNN with ARX and Artificial Neural Network (ANN). The results show that CNN yields RMSE significantly less than that of ARX and ANN. However, the prediction from CNN seems to have lagging behavior in some seasons which might be due to the fluctuation of seasonal weathers.