AI energy modelling & forecasting framework for HVAC

Abstract

Heating, Ventilation, Air-Conditioning systems, or HVACs are known to be one of the highest consumers of electrical power, and this calls the need for energy modelling systems. Deep Learning and AI methods have been recently explored for various practical applications, such as forecasting energy consumption, and this has produced promising outlooks. In this study, Long Short-Term Memory, a Neural-Network Deep Learning algorithm, is used for modelling HVAC energy consumption through historical data. This dataset was first processed through data elimination and transformation, before utilising feature selection tools to determine variables with correlation to energy consumption, then time sequencing is performed. Fine-tuning methods such as hyperparameter tuning and ensemble methods were explored in this study, with an analysis of each method’s impact on the overall predictive performance. Lastly, another dataset is used to test the model’s robustness and adaptability to different data, where the model’s performance was studied for each month, as well as each time sequence.