Edge-based Parametric Digital Twins for Intelligent Building Indoor Climate Modeling

TL;DR

This work addresses the challenge of efficiently forecasting building indoor climate while preserving privacy by deploying an edge-centric integration of parametric digital twins and deep learning. It leverages an ontology-based Brick extension to unify heterogeneous sensor data, and evaluates five DL architectures (LSTM, TCN, TFT, N-HiTS, TiDE) for multi-horizon forecasts on Löfstad Castle using an edge device, achieving strong performance with low inference cost—particularly with the time-series dense encoder TiDE. The study demonstrates that edge deployment reduces latency and data movement, enabling near-real-time insights for HVAC control and maintenance, while highlighting the value of occupancy-aware enhancements. The proposed framework is transferable to manufacturing and other built-environment applications, enabling proactive maintenance and energy optimization in a privacy-preserving, low-latency setting.

Abstract

Digital transformation in the built environment generates vast data for developing data-driven models to optimize building operations. This study presents an integrated solution utilizing edge computing, digital twins, and deep learning to enhance the understanding of climate in buildings. Parametric digital twins, created using an ontology, ensure consistent data representation across diverse service systems equipped by different buildings. Based on created digital twins and collected data, deep learning methods are employed to develop predictive models for identifying patterns in indoor climate and providing insights. Both the parametric digital twin and deep learning models are deployed on edge for low latency and privacy compliance. As a demonstration, a case study was conducted in a historic building in Östergötland, Sweden, to compare the performance of five deep learning architectures. The results indicate that the time-series dense encoder model exhibited strong competitiveness in performing multi-horizon forecasts of indoor temperature and relative humidity with low computational costs.

Figures (8)

• Figure 1: An illustration of the architecture of proposed solution.
• Figure 2: An illustration of multi-horizon forecasting of building indoor climate. In this instance, black dots represent observed values of a particular environmental parameter over a lookback window in the past. Blue triangles are forecasts of the environmental parameter over a forecast horizon in the future. A predictive model uses past observations of a target variable and covariates over the loopback window, along with observations of covariates over the forecast horizon, to make predictions for the target variable.
• Figure 3: Main building of the Löfstad Castle, Östergötland, Sweden.
• Figure 4: Four rooms are selected for comparison. (a) Room 205 on the second floor (2F), (b) Room 103 on the first floor (1F), (c) Room 3 on the ground floor (GF), and (d) Room 05 on the basement floor (BF). The prototype sensor box deployed in Room 103 is depicted in (e). Each of the other three rooms also has been deployed a sensor box that looks like in subfigure (f). Orange dots mark deployment positions. Among the four rooms, only Room 103 has employed electric radiant heating.
• Figure 5: Historical hourly (a) temperature and (b) relative humidity of outdoor and four selected rooms from a beginning time to 23:00 on December 31, 2023. The beginning time is 00:00 on February 14, 2023, for Room 05 and 00:00 on January 13, 2023, for the other three rooms and outdoor. Hours appearing in this paper are expressed in 24-hour format and are all in the timezone of Central European Time.