A Deep Learning Based Digital Twin for Indoor Temperature Prediction in Smart Buildings

With the advent of the Internet of Things (IoT), smart buildings (SBs) are increasingly equipped with a variety of sensors, including those for temperature, humidity, lighting, and human presence. These SBs often incorporate automated heating, ventilation, and air conditioning (HVAC) systems that generate substantial amounts of data daily. This data can be leveraged to develop and optimize energy-saving strategies. In this context, accurate indoor temperature prediction is essential for effective energy management. Moreover, if used properly, this prediction task can also improve the comfort within SBs. This paper proposes a novel approach to predicting indoor temperature called 'Digital Twin for Temperature Prediction (DT4TP)' based on the digital twin paradigm and deep learning (DL) LSTM and BilSTM models. DT4TP has been designed to process real-time and historical data from various IoT sensors in SBs during the winter and summer seasons and to predict the temperature in the future in such environments. DT4TP has been validated through a case study developed at ICAR-CNR in Rende, Italy, and through a state-of-the-art dataset collected from office rooms in Hebei, China. Furthermore, this paper compares the proposed approach with well-known algorithms such as ANN, DNN, CNN, LSTM, and GRU. The comparison uses metrics such as MAE, MSE, RMSE, and R2. The proposed approach can improve the temperature management of the HVAC system offering benefits for SB operations.