A Generative Deep Learning Framework Across Time Series to Optimize the Energy Consumption of Air Conditioning Systems

Working towards active buildings that fully integrate efficient demand management with renewable energy sources and storage, energy efficiency is an important step, as building inefficiencies cause energy wastage and increase energy-related expenses. Currently, static thermal setpoints are typically used to maintain the inside temperature of a building at a comfortable level irrespective of its occupancy. This paper introduces a deep learning framework that trains across time series to forecast the temperatures of a future period directly where a particular room is unoccupied and optimises the setpoints of the room. To the best of our knowledge, this is the first study to use a state-of-the-art deep learning method trained across series to accurately predict temperatures for the subsequent optimal control of room setpoints. In contrast to traditional forecasting approaches that build isolated models to predict each series, our framework uses global recurrent neural network models that are trained with a set of relatively short temperature series, allowing the models to learn cross-series information. The predicted temperatures were then used to define the optimal thermal setpoints to be used inside the room during the unoccupied periods. We evaluate the prediction accuracy of our deep learning framework against a set of state-of-the-art forecasting models and can outperform those by a large margin. Furthermore, we analyse the usage of our deep learning framework to optimise the energy consumption of an air conditioning system in a real-world scenario using temperature data from a university lecture theatre. Based on simulations, we show that our proposed framework can lead to savings of approximately 20% and 15%, respectively, compared to the traditional temperature control model that does not use optimisation techniques and a programmable thermostat.