Engineering-based generic modeling of occupant behavior for energy efficient buildings
Final Report Abstract
The objective of this project was to develop a comprehensive method for modelling energy-related user actions in commercial buildings. In particular, deep learning methods were chosen as the modelling approach. To this end, various feed-forward and Recurrent Neural Network (RNN) architectures were studied and proposed. Particular attention was paid to the applicability of these models to a wide range of occupants and buildings. Considering the state of research at the beginning of the project, the models were initially developed for a sub-case of manually operable windows. The knowledge obtained was then used to model generic occupant actions, occupant behaviour and energy consumption data. The results showed that window states could be modelled effectively using a one-hour time discretization. Furthermore, the inclusion of window state changes over multiple past days could significantly improve both the estimation and prediction accuracy of the 10-minute window state prediction of the feed-forward neural network. In the end, the use of RNN architectures proved effective for modelling both adaptive and non-adaptive actions of the users as well as occupant behavior (OB) and energy consumption data. A key question in this project was how to achieve models’ generalization to different occupants and buildings. The results of sub task 3 showed that domain adaptation is an effective approach that can be conducted effectively using only a small sample of data from the target buildings. Furthermore, the external memory component of the end-to-end memory network of sub task 4 showed to rapidly generalize to the unknown occupants outside of the model’s training distribution. As such, this approach might considerably reduce the number of data samples that are usually needed by OB methods to model the window opening actions. This improved model’s generalization without the need for extensive personal data collection could help to maintain as much data privacy as possible. Finally, the results of sub task 5 showed that the developed miscellaneous energy loads predictive model led to performance improvement in case of evaluation on three independent data sets without any model calibration or domain adaptation, when compared to the current benchmark.
Publications
-
Right on Time - Exploring Suitable Time Discretization for Occupant Behavior Co-Simulation. Building Simulation Conference proceedings, 16, 2099-2106. IBPSA.
Markovic, Romana; Grintal, Eva; Nouri, Amin; Frisch, Jérôme & van, Treeck Christoph
-
Generic occupant behavior modeling for commercial buildings (Doctoral Thesis, RWTH Aachen University).
Markovic, R.
-
Modeling occupant behavior in buildings. Building and Environment, 174, 106768.
Carlucci, Salvatore; De Simone, Marilena; Firth, Steven K.; Kjærgaard, Mikkel B.; Markovic, Romana; Rahaman, Mohammad Saiedur; Annaqeeb, Masab Khalid; Biandrate, Silvia; Das, Anooshmita; Dziedzic, Jakub Wladyslaw; Fajilla, Gianmarco; Favero, Matteo; Ferrando, Martina; Hahn, Jakob; Han, Mengjie; Peng, Yuzhen; Salim, Flora; Schlüter, Arno & van, Treeck Christoph
-
A gap-filling method for room temperature data based on autoencoder neural networks. Building Simulation Conference Proceedings. KU Leuven.
Liguori, Antonio; Markovic, Romana; Frisch, Jérôme; Wagner, Andreas; Causone, Francesco & van, Treeck Christoph
-
Day-ahead prediction of plug-in loads using a long short-term memory neural network. Energy and Buildings, 234, 110667.
Markovic, Romana; Azar, Elie; Annaqeeb, Masab Khalid; Frisch, Jérôme & Treeck, Christoph van
-
Indoor environment data time-series reconstruction using autoencoder neural networks. Building and Environment, 191, 107623.
Liguori, Antonio; Markovic, Romana; Dam, Thi Thu Ha; Frisch, Jérôme; van Treeck, Christoph & Causone, Francesco
-
Prediction of HVAC loads at different spatial resolutions and buildings using deep learning models. Building Simulation Conference Proceedings. KU Leuven.
Liguori, Antonio; Yang, Shiying; Markovic, Romana; Thu, Ha Dam Thi; Wagner, Andreas; Christoph, Christoph & van, Treeck van Treeck
-
A guideline to document occupant behavior models for advanced building controls. Building and Environment, 219, 109195.
Dong, Bing; Markovic, Romana; Carlucci, Salvatore; Liu, Yapan; Wagner, Andreas; Liguori, Antonio; van Treeck, Christoph; Oleynikov, Dmitry; Azar, Elie; Fajilla, Gianmarco; Drgoňa, Ján; Kim, Joyce; Vellei, Marika; De Simone, Marilena; Shamsaiee, Masood; Bavaresco, Mateus; Favero, Matteo; Kjaergaard, Mikkel; Osman, Mohamed ... & Kang, Xuyuan
