Black-Box Model for Predicting HVAC Energy Consumption and PCM State with Different Melting Temperatures of PCM for Small Office Buildings

Heating, Ventilation, and Air-Conditioning (HVAC) operations consume around 40% of buildings' energy consumption (EC). To reduce this consumption, Phase change material (PCM) can be incorporated in buildings' envelopes. However, the effectiveness of PCM varies depending on its thermophysical properties and climate conditions. Conventionally, finding the optimal PCM properties requires a computationally extensive parametric study of white-box model, which may not always provide accurate results due to input simplification. To overcome this shortcoming, machine learning (ML) algorithms can be developed to identify optimal PCM properties based on predicting HVAC loads and PCM state. The best performing models for predicting cooling and heating energy consumptions were achieved by Artificial Neural Network (ANN) with MAE around 8% and Random Forest (RF) with MAE around 45%, respectively. As for classification, RF with Polynomial kernel PCA showed an accuracy of 81.5%. These results demonstrate the potential of ML algorithms to optimize building energy efficiency.